https://reading-list.oddship.net A curated linklog of essays, posts, papers, and notes. Wed, 26 Aug 2026 21:45:00 +0530 https://reading-list.oddship.net/notes/2026-08-26-ducklabs-joins-aws/ https://reading-list.oddship.net/notes/2026-08-26-ducklabs-joins-aws/ Wed, 26 Aug 2026 21:45:00 +0530 Logged at IST: 2026-08-26 21:45 IST What it is: DuckLabs' announcement that it is joining AWS, shared by Frank Denis. Gist: DuckLabs says the Amsterdam team behind DuckDB will stay together under AWS, continuing work on DuckDB, DuckLake, Quack, and the wider Duck Stack. The company is presenting this as a scale move: DuckDB now has more than one million downloads a day, and DuckLabs says AWS gives it the resources, infrastructure, and customer reach to grow without turning the small team into a large sales and support organization. The trust question is the important part. DuckLabs says DuckDB… https://reading-list.oddship.net/notes/2026-08-26-go-sync-map-hash-trie/ https://reading-list.oddship.net/notes/2026-08-26-go-sync-map-hash-trie/ Wed, 26 Aug 2026 16:33:00 +0530 Logged at IST: 2026-08-26 16:33 IST What it is: Phuong Le's VictoriaMetrics walkthrough of Go's hash-trie-backed sync.Map implementation. Gist: The visible API is still the familiar Load, Store, Delete, LoadOrStore, CompareAndSwap, Range, and Clear. Underneath, Go moved sync.Map to an internal HashTrieMap: keys are hashed, the hash is consumed in 4-bit chunks, each indirect node has 16 child slots, and full hash collisions fall back to overflow entries. The concurrency model is the useful part. Load can walk atomic child pointers without taking node mutexes. Writers search without a lock, then… https://reading-list.oddship.net/notes/2026-08-26-go-generic-methods-interface-boundary/ https://reading-list.oddship.net/notes/2026-08-26-go-generic-methods-interface-boundary/ Wed, 26 Aug 2026 10:14:00 +0530 Logged at IST: 2026-08-26 10:14 IST What it is: Dominik Honnef's explanation of generic methods in Go 1.27. Gist: Go 1.27 removes a long-standing awkwardness in Go generics: methods on concrete types can now declare their own type parameters. That means an API like Node[T].Map[U] no longer has to pollute the receiver type with a U that only one method needs, or move the operation out into a package-level helper function. The interesting part is the boundary Go keeps. Interfaces still cannot declare generic methods, and a concrete generic method does not satisfy a matching interface method. Hon… https://reading-list.oddship.net/notes/2026-08-26-persona-webmcp-chat-widget/ https://reading-list.oddship.net/notes/2026-08-26-persona-webmcp-chat-widget/ Wed, 26 Aug 2026 10:01:00 +0530 Logged at IST: 2026-08-26 10:01 IST What it is: Persona is Runtype's framework-free TypeScript chat widget for adding agentic front-end experiences to websites, with explicit WebMCP/page-tool support. Gist: The repo positions Persona as a small, pluggable, themeable AI chat UI that can be dropped into existing sites with vanilla JS or bundled from npm. It supports streaming responses, voice I/O, multimodal attachments, tool-call visualization, approval gates, artifacts, local client tools, and WebMCP/page tools. The timing matters because the shared post was a reply to OpenAI's WebMCP desktop-… https://reading-list.oddship.net/notes/2026-08-26-paul-dix-end-of-programming/ https://reading-list.oddship.net/notes/2026-08-26-paul-dix-end-of-programming/ Wed, 26 Aug 2026 09:53:00 +0530 Logged at IST: 2026-08-26 09:53 IST What it is: Paul Dix's argument that frontier agentic coding has crossed from autocomplete into software-factory territory. Gist: Dix uses Bun 1.4's million-line Zig-to-Rust rewrite as the headline case: one developer built the harness, agents did the translation and follow-on refinement, and the result shipped to millions of developer machines. His point is not just that AI can make prototypes faster. It is that with a good oracle, enough tokens, and a tight verification loop, agents can produce and harden large amounts of working software. He connects that… https://reading-list.oddship.net/notes/2026-08-25-go-json-v2-migration-guide/ https://reading-list.oddship.net/notes/2026-08-25-go-json-v2-migration-guide/ Tue, 25 Aug 2026 20:19:00 +0530 Logged at IST: 2026-08-25 20:19 IST What it is: Anton Zhiyanov's interactive guide to Go's new json/v2 package. Gist: The guide is a migration map from encoding/json v1 to json/v2. The basics still look familiar, but the surrounding API changes a lot: direct MarshalWrite and UnmarshalRead, streaming encode/decode through jsontext, composable options instead of many one-off functions, new struct tags like case and embed, and streaming custom marshaling through MarshalJSONTo / UnmarshalJSONFrom. The defaults also change in ways that will matter in real services. Nil slices encode as [] instead o… https://reading-list.oddship.net/notes/2026-08-25-go-llama-pure-go-local-inference/ https://reading-list.oddship.net/notes/2026-08-25-go-llama-pure-go-local-inference/ Tue, 25 Aug 2026 20:15:00 +0530 Logged at IST: 2026-08-25 20:15 IST What it is: goccy/go-llama, a pure-Go llama.cpp-style inference library for running GGUF models from Go applications. Gist: The interesting bit is the packaging path. go-llama does not bind to llama.cpp through cgo, and it does not embed a wasm runtime. Instead, llama.cpp is compiled to WASI WebAssembly, then translated ahead-of-time into standalone Go through wasm2go and llamawasm2go. The result is meant to be a normal Go dependency: no shared library, no cgo, and static-binary friendly. The README claims a fairly complete local-inference surface: model/con… https://reading-list.oddship.net/notes/2026-08-25-headlong-persistent-agent-microharness/ https://reading-list.oddship.net/notes/2026-08-25-headlong-persistent-agent-microharness/ Tue, 25 Aug 2026 13:06:00 +0530 Logged at IST: 2026-08-25 13:06 IST What it is: Laude's launch post for Headlong, an open-source Bash microharness for persistent agents. Gist: Headlong pushes past the usual reactive agent model. Instead of waiting for a task, doing it, and freezing again, a Headlong agent keeps generating thoughts in a continuous loop. Human messages from Slack, Telegram, or the web UI land as observations in that single thought stream, and the agent decides if and when to reply. The implementation is intentionally small: less than 10K lines of Bash in the core, shellm as a recursive-language-model loop, tra… https://reading-list.oddship.net/notes/2026-08-25-speculative-programmatic-tool-calling/ https://reading-list.oddship.net/notes/2026-08-25-speculative-programmatic-tool-calling/ Tue, 25 Aug 2026 12:48:00 +0530 Logged at IST: 2026-08-25 12:48 IST What it is: Alex Zhang's post and reference implementation for Speculative Programmatic Tool Calling, or sPTC. Gist: sPTC is a latency trick for agents where the model's main action is code in a REPL. Instead of waiting for the whole generated program to finish before running tools, the harness watches the code stream, speculatively parses likely tool calls, launches expensive sub-agent or sub-LLM calls early, and then lets the real execution claim those cached futures if the calls actually happen. The idea matters because programmatic tool calling makes the… https://reading-list.oddship.net/notes/2026-08-25-private-agent-dev-repo/ https://reading-list.oddship.net/notes/2026-08-25-private-agent-dev-repo/ Tue, 25 Aug 2026 12:35:00 +0530 Logged at IST: 2026-08-25 12:35 IST What it is: Ryan Dahl describing how celld is developed from a larger private denoland/celld.dev repo, while the public denoland/celld repo stays a lean export. Gist: The useful part is the shape of the working repo. Dahl says the private repo has an LLM-maintained design wiki, a large test suite, a TLA+ spec, website code, cloud experiment scripts, and a speculative managed control plane. The public repo is open source, but the real development loop happens around the richer private artifact. That feels like an agent-era pattern: code is no longer the only … https://reading-list.oddship.net/notes/2026-08-23-find-problems-staff-engineer/ https://reading-list.oddship.net/notes/2026-08-23-find-problems-staff-engineer/ Sun, 23 Aug 2026 18:21:00 +0530 Logged at IST: 2026-08-23 18:21 IST What it is: Lalit Maganti on how he finds problems worth solving as a Staff Engineer, especially in infrastructure and developer-tools contexts. Gist: The useful move is to treat problem discovery as ambient listening, not a scheduled “think strategically” exercise. Maganti watches the normal flow of meetings, chats, emails, and complaints, then asks follow-up questions to understand the real workflow pain behind requested solutions. He is careful not to jump on the first loud request. Problems need to accumulate: the same pain showing up across teams is str… https://reading-list.oddship.net/notes/2026-08-23-fast-hard-code/ https://reading-list.oddship.net/notes/2026-08-23-fast-hard-code/ Sun, 23 Aug 2026 11:13:00 +0530 Logged at IST: 2026-08-23 11:13 IST What it is: Armin Ronacher's short post on how LLMs change language and technology choices when starting new projects. Gist: Ronacher's point is not that programming is solved. It is that the cost of becoming productive in an unfamiliar language has dropped enough that old language-choice friction matters less. If an agent can draft, translate, and optimize code in a language I barely know, I can choose more for project identity, performance goals, and the promise of small, fast software. That changes which technologies feel approachable. Rust and Zig benefi… https://reading-list.oddship.net/notes/2026-08-21-deepseek-v4-flash-demand-capacity/ https://reading-list.oddship.net/notes/2026-08-21-deepseek-v4-flash-demand-capacity/ Fri, 21 Aug 2026 10:12:00 +0530 Logged at IST: 2026-08-21 10:12 IST What it is: Jay from OpenCode explaining what happened after DeepSeek v4 Flash became the cheap default for a lot of coding-agent traffic. Gist: The useful part is the operator view of a model getting cheap enough to change user behavior. Jay says DeepSeek v4 Flash launched on August 1 and grew on OpenCode from roughly 3T tokens/day to 18T tokens/day in two weeks, close to doubling OpenRouter's daily volume and possibly 30 to 50% of DeepSeek's own volume. The cause, in his telling, was not just model quality. It was price. Users got a first taste of AI that … https://reading-list.oddship.net/notes/2026-08-20-typescript-go-content-mappers/ https://reading-list.oddship.net/notes/2026-08-20-typescript-go-content-mappers/ Thu, 20 Aug 2026 18:37:00 +0530 Logged at IST: 2026-08-20 18:37 IST What it is: Microsoft’s typescript-go PR #4712, “Content mappers”, merged by Andrew Branch and called out as likely to land in TypeScript 7.1. Gist: Content mappers let TypeScript include otherwise unsupported file types in a program. A mapper turns a foreign file such as .vue, .svelte, .astro, or an Angular-style source into valid JS, JSX, TS, TSX, or JSON, then returns span mappings back to the original content. The PR defines the operational contract around that idea. Projects can declare mapper packages in tsconfig.json, mapper packages describe their ex… https://reading-list.oddship.net/notes/2026-08-20-rust-arrayref-supply-chain-attack/ https://reading-list.oddship.net/notes/2026-08-20-rust-arrayref-supply-chain-attack/ Thu, 20 Aug 2026 17:58:00 +0530 Logged at IST: 2026-08-20 17:58 IST What it is: A Rust Security Response Team advisory about a supply-chain attack involving arrayref and several malicious crates on crates.io. Gist: Rust's security-response team says it received a report at 2026-08-20 07:15 UTC that the proc-macro1 crate was malicious. The team verified that the crate's build script downloaded a malicious payload, then deleted proc-macro1 and related crates including proc-macro-en, aovine, arone, aronenao, and tinymember. The incident also touched real ecosystem packages. arrayref@0.3.10 had been republished to depend on the … https://reading-list.oddship.net/notes/2026-08-20-sqlite-in-nix-wasm/ https://reading-list.oddship.net/notes/2026-08-20-sqlite-in-nix-wasm/ Thu, 20 Aug 2026 11:17:00 +0530 Logged at IST: 2026-08-20 11:17 IST What it is: Farid Zakaria's write-up on trying to expose SQLite-backed queries to Nix for nixpkgs-multiverse. Gist: nixpkgs-multiverse currently carries JSON indexes, but builtins.fromJSON is eager: one lookup pays the cost of parsing and materializing the whole file. Zakaria uses that as the setup for three SQLite-in-Nix experiments. builtins.exec is the simple unsafe escape hatch: ask sqlite3 to print a Nix expression and let the evaluator parse it. builtins.importNative goes further by loading a shared object that can cache SQLite handles and build Nix va… https://reading-list.oddship.net/notes/2026-08-20-go-1-27-release/ https://reading-list.oddship.net/notes/2026-08-20-go-1-27-release/ Thu, 20 Aug 2026 06:31:00 +0530 Logged at IST: 2026-08-20 06:31 IST What it is: The official Go 1.27 release post, shared through mattn's Japanese X thread of highlights. Gist: Go 1.27 is a practical release across the language, toolchain, runtime, and standard library. The language headline is generic methods, alongside broader generic function inference and direct promoted-field keys in struct literals. The operational pieces are just as interesting. go fix gets more modernizers, go doc can query package@version, go mod tidy normalizes require blocks, small object allocation gets faster, and the goroutine leak profile is n… https://reading-list.oddship.net/notes/2026-08-19-rangebitmap-slicez-kubernetes-metadata/ https://reading-list.oddship.net/notes/2026-08-19-rangebitmap-slicez-kubernetes-metadata/ Wed, 19 Aug 2026 14:28:00 +0530 Logged at IST: 2026-08-19 14:28 IST What it is: Richard Startin points to a Yandex Monium write-up using RangeBitmap, a range-encoded bit-sliced index he designed for Apache Pinot and which is available in RoaringBitmap. His follow-up links SliceZ, his newer Java successor for range and point queries over unsorted numeric data. Gist: Yandex Monium stores billions of metrics, processes about 50 GB/s of logs, and serves around 16,000 internal users. Kubernetes broke an old metadata assumption: pod labels are short-lived, but historical label values accumulate. After a few weeks, selectors can hi… https://reading-list.oddship.net/notes/2026-08-19-ai-removing-middle-class-software-engineering/ https://reading-list.oddship.net/notes/2026-08-19-ai-removing-middle-class-software-engineering/ Wed, 19 Aug 2026 11:10:00 +0530 Logged at IST: 2026-08-19 11:10 IST What it is: Florian Herrengt's essay arguing that AI is removing the middle class of software engineering. Gist: The core claim is not "AI is bad". Herrengt says he uses AI heavily. The warning is that AI removes the speed limit from weak engineering culture. Before AI, bad decisions compounded slowly because implementation was expensive. Now someone can generate a 25,000-line PR, add services, change schemas, and ship plausible-looking functionality before anyone has built a real mental model of what changed. The output may work at first, but the debt is hi… https://reading-list.oddship.net/notes/2026-08-19-git-at-any-scale-cursor/ https://reading-list.oddship.net/notes/2026-08-19-git-at-any-scale-cursor/ Wed, 19 Aug 2026 09:51:00 +0530 Logged at IST: 2026-08-19 09:51 IST What it is: Vicent Martí's Cursor post on the storage system behind Origin, Cursor's Git hosting product. Gist: The useful framing is that Git hosting is hard because Git was designed so every repository copy is equivalent. That makes the local developer workflow great, but it makes server-side scaling awkward: packfiles are optimized for local disk and Git clients still expect packfiles over the network. The post walks through prior approaches. Object-level distributed stores map nicely to content-addressed Git objects, but DAG walks become round-trip-heavy… https://reading-list.oddship.net/notes/2026-08-19-postgres-poisoned-connection-pools/ https://reading-list.oddship.net/notes/2026-08-19-postgres-poisoned-connection-pools/ Wed, 19 Aug 2026 09:33:00 +0530 Logged at IST: 2026-08-19 09:33 IST What it is: PlanetScale's debugging note on poisoned Postgres connection pools, shared by Ben Dicken. Gist: The failure mode is simple and nasty: PgBouncer transaction pooling reuses underlying database connections across clients. If one client leaves session state behind, the next client can inherit it. The concrete example is read-only state. Application code used SET SESSION CHARACTERISTICS AS TRANSACTION READ ONLY for a read-only transaction, but that changed the session rather than only the transaction. Later requests reused that same underlying connect… https://reading-list.oddship.net/notes/2026-08-18-ai-plays-age-of-empires-ii/ https://reading-list.oddship.net/notes/2026-08-18-ai-plays-age-of-empires-ii/ Tue, 18 Aug 2026 16:47:00 +0530 Logged at IST: 2026-08-18 16:47 IST What it is: Emergent Garden's video on using modern AI agents to play Age of Empires II, along with the linked AgentsOfEmpires runner repo. Gist: The video description says the creator used LLM-powered agents, including Claude, GPT, Gemini, and Kimi K3, to write custom Age of Empires II AI scripts, make them fight, and then tune strategies over repeated runs. The listed chapters move from "Agents of Empires" into playing the game, model comparisons, strategy optimization, and battle optimization. The linked GitHub repo is useful context because it shows the … https://reading-list.oddship.net/notes/2026-08-18-detecting-goroutine-leaks/ https://reading-list.oddship.net/notes/2026-08-18-detecting-goroutine-leaks/ Tue, 18 Aug 2026 14:15:00 +0530 Logged at IST: 2026-08-18 14:15 IST What it is: Anton Zhiyanov's walkthrough of goroutine leak detection in modern Go, comparing goleak, synctest, and the experimental goroutineleak pprof profile. Gist: The article starts with the practical shape of a goroutine leak: goroutines blocked forever on synchronization primitives while the rest of the program keeps running. That makes leaks quieter than deadlocks and harder to catch than data races. Zhiyanov walks through a simple unbuffered-channel example, then compares three ways to find the leak. goleak works today as a third-party test helper. s… https://reading-list.oddship.net/notes/2026-08-17-software-cartographers/ https://reading-list.oddship.net/notes/2026-08-17-software-cartographers/ Mon, 17 Aug 2026 11:37:00 +0530 Logged at IST: 2026-08-17 11:37 IST What it is: Adam Craven's 2023 essay arguing for "software cartographers": people and processes that keep maps of working software in sync with the code. Gist: The essay uses the village, town, and city analogy to explain why small codebases fit inside one engineer's head while large systems stop being knowable by any individual, or eventually by the whole team. Other complex human-built systems use maps to reduce complexity. Software usually has diagrams at the start, then lets them rot once code becomes the only source of truth. The practical claim is that… https://reading-list.oddship.net/notes/2026-08-16-softer-software/ https://reading-list.oddship.net/notes/2026-08-16-softer-software/ Sun, 16 Aug 2026 13:20:00 +0530 Logged at IST: 2026-08-16 13:20 IST What it is: plugyawn's essay arguing that LLM prompts and agent harnesses can be understood as a softer generalization of software. Gist: The piece starts from SICP's definition of a program as a precise description of a process, then asks whether software can have the same kind of relaxed version that hard functions get in softmax or other continuous approximations. The proposed answer is that an LLM is a stochastic interpreter for informal language. A prompt describes a computational process, the model searches programspace, and the output is the execution… https://reading-list.oddship.net/notes/2026-08-14-deepseek-harness-pi-cross-pollination/ https://reading-list.oddship.net/notes/2026-08-14-deepseek-harness-pi-cross-pollination/ Fri, 14 Aug 2026 22:56:00 +0530 Logged at IST: 2026-08-14 22:56 IST What it is: Tianyi Cui from DeepSeek quoted Armin Ronacher's reaction to DeepSeek Harness and said DeepSeek Harness reused Pi's LLM adapter package for connecting to non-DeepSeek models. Gist: The tweet is small, but the ecosystem signal is good. Armin says DeepSeek Harness is not perfect, but it is the first new thing in the space that made him feel inspired to revisit some of Pi/OMP's choices. Tianyi says Pi is a daily driver for many DeepSeek researchers and developers, and that DSH reused Pi's LLM adapter package for non-DeepSeek models. The underlying s… https://reading-list.oddship.net/notes/2026-08-14-miren-rpc-that-calls-you-back/ https://reading-list.oddship.net/notes/2026-08-14-miren-rpc-that-calls-you-back/ Fri, 14 Aug 2026 10:05:00 +0530 Logged at IST: 2026-08-14 10:05 IST What it is: Miren engineering post by Evan Phoenix and Paul Hinze on the RPC layer they built for typed two-way calls across CLI, control plane, and runtime components. Gist: Miren argues that a lot of deployment-platform work does not fit cleanly into one request and one response. A deploy has progress updates, health checks, subscriptions, and callbacks from a server to an ephemeral client. Polling, webhooks, or a separate stream work, but they make the request/response model leak. Their answer is object-capability RPC. In the IDL, parameters and results c… https://reading-list.oddship.net/notes/2026-08-13-go-sync-nocopy/ https://reading-list.oddship.net/notes/2026-08-13-go-sync-nocopy/ Thu, 13 Aug 2026 17:56:00 +0530 Logged at IST: 2026-08-13 17:56 IST What it is: func25.dev explaining Go's internal sync.noCopy marker and how go vet detects accidental copies of sync-like structs. Gist: noCopy is not compiler magic. It is an empty marker with two empty methods on its pointer receiver: Lock and Unlock. Those names make *noCopy satisfy sync.Locker, while the value noCopy does not. go vet's copylocks checker uses that shape: while inspecting copied types, it recursively looks for a type whose pointer implements sync.Locker but whose value does not. That is why go build will still accept copying a used sync.Map… https://reading-list.oddship.net/notes/2026-08-13-appwrite-cli-go-rewrite/ https://reading-list.oddship.net/notes/2026-08-13-appwrite-cli-go-rewrite/ Thu, 13 Aug 2026 09:59:00 +0530 Logged at IST: 2026-08-13 09:59 IST What it is: Richard Seroter pointing to Appwrite's rewrite of its CLI from TypeScript/Bun to Go. Gist: Appwrite says the user-facing contract stays the same: npm package name, appwrite binary, flags, scripts, CI behavior, exit codes, and JSON output. The internals changed from a JavaScript bundle on a runtime plus 189 packages underneath to a native Go binary. The measured deltas are the useful part: startup went from 207.6ms to 11.1ms, install dependency count from 330 packages to 2, on-disk install from 209MB to 13MB, memory during push from 283.5MB to 28.… https://reading-list.oddship.net/notes/2026-08-12-html-over-wire-sse-websockets/ https://reading-list.oddship.net/notes/2026-08-12-html-over-wire-sse-websockets/ Wed, 12 Aug 2026 19:04:00 +0530 Logged at IST: 2026-08-12 19:04 IST What it is: A YAGNI Club response to an “HTML over WebSockets” article, arguing that many HTML-over-the-wire apps should consider SSE plus normal POSTs before reaching for WebSockets. Gist: The core correction is transport-level. The original WebSocket argument says one persistent connection avoids TCP and HTTP overhead, but with HTTP/2 keep-alive the page, CSS, SSE stream, and POST commands can already share the same connection. A WebSocket often means a second TCP/TLS connection plus an HTTP/1.1 upgrade, while command POSTs keep the normal HTTP request lif… https://reading-list.oddship.net/notes/2026-08-12-super-mario-rl-agent/ https://reading-list.oddship.net/notes/2026-08-12-super-mario-rl-agent/ Wed, 12 Aug 2026 18:56:00 +0530 Logged at IST: 2026-08-12 18:56 IST What it is: Shantanu Goel’s writeup on training a PPO reinforcement-learning agent, using stable-retro and stable-baselines3, to beat World 1-1 of NES Super Mario Bros. Gist: The interesting part is the debugging path. Early attempts failed because the agent saw only stacked 84×84 grayscale frames, produced too-short jumps, and treated three Mario lives as one long episode. Moving to one-life episodes helped, but the bigger fixes were reward shaping and observation design: dense rewards for forward progress, coins, score, time pressure, and a modest flag bon… https://reading-list.oddship.net/notes/2026-08-12-ai-software-steak-machine/ https://reading-list.oddship.net/notes/2026-08-12-ai-software-steak-machine/ Wed, 12 Aug 2026 10:27:00 +0530 Logged at IST: 2026-08-12 10:27 IST What it is: Joe Ingeno sharing Yurii Sydorets’ essay, “Almost No Skill Required to Cook a Steak (Though You Probably Can’t Make a Decent One).” Gist: Sydorets uses the steak analogy well: AI can make software creation feel as easy as putting meat in a hot pan, but consistently good results are still a craft problem. Models can follow recipes, generate starting points, and automate repetitive work, but they do not know the taste in your head unless you translate it into requirements, constraints, examples, tests, and feedback. The sharper point is that outsou… https://reading-list.oddship.net/notes/2026-08-11-go-ai-assisted-software-engineering/ https://reading-list.oddship.net/notes/2026-08-11-go-ai-assisted-software-engineering/ Tue, 11 Aug 2026 23:16:00 +0530 Logged at IST: 2026-08-11 23:16 IST What it is: Google Developers Blog essay by Cameron Balahan and Richard Seroter arguing that Go is well-suited to AI-assisted software engineering. Gist: The essay’s core frame is that AI shifts the bottleneck from writing code to reviewing, verifying, and maintaining code. In that world, Go’s long-standing bias toward team-scale software engineering becomes more valuable: a strict compiler, gofmt, a unified toolchain, standard testing, dependency management, vulnerability scanning, and compatibility guarantees give both humans and agents deterministic guard… https://reading-list.oddship.net/notes/2026-08-11-spotify-xirp-agentic-development-environment/ https://reading-list.oddship.net/notes/2026-08-11-spotify-xirp-agentic-development-environment/ Tue, 11 Aug 2026 18:44:00 +0530 Logged at IST: 2026-08-11 18:44 IST What it is: Spotify Engineering's launch of Xirp, a vendor-neutral agentic development environment that works across Claude, Gemini, and OpenAI Codex. Gist: Spotify frames Xirp as an agent harness with institutional memory. The pitch is that coding tools have improved generation, but agents still make locally correct and operationally wrong choices when they lack service ownership, dependency, docs, and architectural-decision context. The product page says Xirp connects to Spotify Portal so agents and engineers share a context layer: services, owners, depend… https://reading-list.oddship.net/notes/2026-08-11-deepseek-v4-flash-agent-harnesses/ https://reading-list.oddship.net/notes/2026-08-11-deepseek-v4-flash-agent-harnesses/ Tue, 11 Aug 2026 09:50:00 +0530 Logged at IST: 2026-08-11 09:50 IST What it is: Composio's benchmark image comparing DeepSeek V4 Flash on four agent harnesses: Pi Agent, Prime Agent, Deep Agents, and Hermes Agent. Gist: The post says Composio ran DeepSeek V4 Flash through 30 challenging agentic tasks on four more harnesses. The headline result is that Pi Agent had the best reported combination: 66.7% pass rate, $0.012 median cost per task, and 132s median time per task. The chart's other reported figures are Prime Agent at 62.5% pass, $0.045, 242s; Deep Agents at 53.3%, $0.018, 187s; and Hermes Agent at 50.0%, $0.017, 176s. … https://reading-list.oddship.net/notes/2026-08-10-source-code-availability/ https://reading-list.oddship.net/notes/2026-08-10-source-code-availability/ Mon, 10 Aug 2026 16:19:00 +0530 Logged at IST: 2026-08-10 16:19 IST What it is: Loris Cro’s long essay on source-code availability, using Zine, Zig, Radicle, Codeberg, Tangled, community mirrors, and package registries as examples. Gist: Cro starts from a practical build problem: if a project’s dependencies live across GitHub, Codeberg, and small self-hosted forges, fresh builds can fail whenever any host is down. Forking and vendoring solve this locally, but they do not make the copied code discoverable or automatically useful as shared redundancy. The core argument is that source availability is a cost-allocation problem. … https://reading-list.oddship.net/notes/2026-08-10-endless-execution/ https://reading-list.oddship.net/notes/2026-08-10-endless-execution/ Mon, 10 Aug 2026 11:13:00 +0530 Logged at IST: 2026-08-10 11:13 IST What it is: DHH’s short HEY World post about the experience of using agents as endless execution. Gist: DHH argues that the age of agents has made every idea, hunch, and experiment feel immediately reachable. His claim is intentionally experiential rather than cautious: after decades of loving computers, the ability to execute almost any passing idea is the most fun he has had with one. The useful signal is not that agents are perfect or risk-free. It is that the interface between imagination and execution has changed. For people with a surplus of ideas, age… https://reading-list.oddship.net/notes/2026-08-08-logchef-2-open-source-log-analytics/ https://reading-list.oddship.net/notes/2026-08-08-logchef-2-open-source-log-analytics/ Sat, 08 Aug 2026 19:00:00 +0530 Logged at IST: 2026-08-08 19:00 IST What it is: Karan Sharma's launch thread for Logchef 2.0, an open-source, self-hosted log analytics workspace for ClickHouse and VictoriaLogs. Gist: Logchef 2.0 expands the project from a fast ClickHouse log explorer into a broader operational workspace. The new shape is search, live tail, dashboards, alerts, RBAC, CLI, AI query assistant, and MCP integration in one self-hosted binary. The product choice I like here is that Logchef does not try to become another storage layer. It queries ClickHouse and VictoriaLogs directly, so existing ingestion, retention,… https://reading-list.oddship.net/notes/2026-08-08-ai-adoption-is-a-myth/ https://reading-list.oddship.net/notes/2026-08-08-ai-adoption-is-a-myth/ Sat, 08 Aug 2026 16:05:00 +0530 Logged at IST: 2026-08-08 16:05 IST What it is: Mario Zechner recommends Vas's X Article, "AI Adoption is a Myth", saying he has seen similar patterns at smaller scale. Gist: Vas argues that enterprise AI adoption metrics hide a barbell. In his telling, a rollout can produce 5-10% power users, roughly 20% weak users, and a large majority who barely use the tool at all. The dashboard still says adoption happened, but the organization does not get faster. The stronger point is that using AI well is a craft, not a login event. The useful user knows when to clear context, when to turn repeated wor… https://reading-list.oddship.net/notes/2026-08-08-arc-prize-deepseek-v4-flash-0731/ https://reading-list.oddship.net/notes/2026-08-08-arc-prize-deepseek-v4-flash-0731/ Sat, 08 Aug 2026 11:56:00 +0530 Logged at IST: 2026-08-08 11:56 IST What it is: ARC Prize's verified result page for DeepSeek V4 Flash 0731, plus the Hacker News discussion around it. Gist: ARC Prize reports DeepSeek V4 Flash 0731 at max effort scoring 89.0% on ARC-AGI-1 Semi-Private at about $0.02 per task and 61.4% on ARC-AGI-2 Semi-Private at about $0.04 per task. The high and low reasoning variants step down to 87.0%/56.0% and 84.0%/46.0%. That makes the result interesting as a cost-to-capability marker. The HN discussion is mostly reading it as a practical threshold moment: not necessarily frontier SOTA, but cheap enoug… https://reading-list.oddship.net/notes/2026-08-07-exe-dev-software-factory-inventory/ https://reading-list.oddship.net/notes/2026-08-07-exe-dev-software-factory-inventory/ Fri, 07 Aug 2026 16:17:00 +0530 Logged at IST: 2026-08-07 16:17 IST What it is: exe.dev’s X post pointing to Philip Zeyliger’s inventory of the internal agents, bots, and operational tools they use to run exe. Gist: The post is useful because it lists the actual places where exe.dev has put agents and small internal systems into its operating loop. They have agents for systematic security review, alert investigation, daily log-trend emails, and deploy supervision. They also have bots that look for flaky or slow tests, a homegrown status page, Pushover-based phone paging, and daily Slack reports about git commits and support/… https://reading-list.oddship.net/notes/2026-08-07-end-of-no-code-agentic-linux/ https://reading-list.oddship.net/notes/2026-08-07-end-of-no-code-agentic-linux/ Fri, 07 Aug 2026 03:03:00 +0530 Logged at IST: 2026-08-07 03:03 IST What it is: David Crawshaw’s X post saying “the age of no code has passed,” pointing to Philip Zeyliger / exe.dev’s essay “The End of No Code.” Gist: The essay argues that no-code and low-code platforms solved a real organizational problem: business teams needed better structure than spreadsheets, but buying or deploying custom software was blocked by procurement, IT, infrastructure, or engineering scarcity. Airtable’s core insight was that letting users express tables and data types directly lets them model their business without writing ALTER TABLE stateme… https://reading-list.oddship.net/notes/2026-08-06-hypeapps-ham-flutter-markup/ https://reading-list.oddship.net/notes/2026-08-06-hypeapps-ham-flutter-markup/ Thu, 06 Aug 2026 19:45:00 +0530 Logged at IST: 2026-08-06 19:45 IST What it is: HypeApps / HAM, a Flutter-backed approach for rendering HTML-like markup as real native mobile UI, paired with the hypermedia-tv video “Worlds easiest mobile app framework.” It is not open source yet: the FAQ says licensing and distribution have not been announced, and the standalone SDK is planned for Fall 2026. Gist: HypeApps is in the Hyperview/server-driven-UI family, but swaps in Flutter as the renderer and an HTML-like authoring model called HAM. Your server returns markup; the app parses it and renders real native widgets. The server can c… https://reading-list.oddship.net/notes/2026-08-06-hobby-programming-llms-craft/ https://reading-list.oddship.net/notes/2026-08-06-hobby-programming-llms-craft/ Thu, 06 Aug 2026 18:52:00 +0530 Logged at IST: 2026-08-06 18:52 IST What it is: Fogus’s short essay on why niche hobby programming communities are hostile to LLM-generated development, paired with the Hacker News discussion, the linked Coda chess-engine GitHub issue, and Rohan Verma’s adjacent essay on hobbyist programmers being pressured to turn computing into career, monetization, or public validation. Gist: Fogus’s operative claim is concrete: in communities like OSDev, LangDev, TxtDev, EmuDev, RLDev, the demoscene, chess engines, and code golf, the process of mastering a difficult field is often the product. A working pr… https://reading-list.oddship.net/notes/2026-08-06-go-performance-profiling-loop/ https://reading-list.oddship.net/notes/2026-08-06-go-performance-profiling-loop/ Thu, 06 Aug 2026 18:30:00 +0530 Logged at IST: 2026-08-06 18:30 IST What it is: Aliaksandr Valialkin’s X conversation chain about Go, C, Rust, and practical optimization, with context from Go and VictoriaMetrics profiling docs. Gist: The chain starts with a deliberately provocative claim: “Go is faster than Rust.” Valialkin’s actual argument is about the whole optimization loop, not a universal language benchmark. He points to two practical advantages: every Go program can expose low-overhead CPU and memory profiling, and Go rebuilds large services such as VictoriaMetrics in seconds rather than minutes. The replies then forc… https://reading-list.oddship.net/notes/2026-08-05-celld-self-hosted-durable-objects/ https://reading-list.oddship.net/notes/2026-08-05-celld-self-hosted-durable-objects/ Wed, 05 Aug 2026 20:19:00 +0530 Logged at IST: 2026-08-05 20:19 IST What it is: Ryan Dahl's launch post for celld, a Deno open-source daemon for self-hosted, distributed Durable Objects and Workers. Gist: celld aims to run the Cloudflare Workers and Durable Objects programming model on your own machines. Each object is its own SQLite database, addressed by name and replicated to an S3-compatible bucket. The fleet coordinates through that bucket alone, using object-storage compare-and-swap so one node owns a cell at a time without a separate control plane, membership protocol, failure detector, or consensus service. The tweet… https://reading-list.oddship.net/notes/2026-08-05-cloudflare-os/ https://reading-list.oddship.net/notes/2026-08-05-cloudflare-os/ Wed, 05 Aug 2026 19:44:00 +0530 Logged at IST: 2026-08-05 19:44 IST; updated 2026-08-05 19:56 IST with Kenton Varda's launch thread and the GitHub README. What it is: Cloudflare's announcement of Cloudflare OS, an open-source platform for organization-scoped agents, apps, and internal workflows. Gist: Cloudflare OS is pitched as a company-specific work environment rather than a generic chatbot. It combines Cloudflare Access, AI Gateway, Gatekeepers, MCP Server Portals, Skills, Workflows, and generated apps so employees can automate work and reach internal systems under policy. Kenton Varda's launch thread gives the sharper f… https://reading-list.oddship.net/notes/2026-08-05-every-byte-matters/ https://reading-list.oddship.net/notes/2026-08-05-every-byte-matters/ Wed, 05 Aug 2026 08:44:00 +0530 Logged at IST: 2026-08-05 08:44 IST What it is: Farid Zakaria's post on how struct size, cache lines, and working-set size affect performance. Gist: The post starts from a common blind spot: in large Java-style classes, adding another field rarely feels like a performance decision. Zakaria shows why it is one. A CPU cache line is 64 bytes on his machine, so reading one byte from memory often pulls the surrounding 64 bytes into cache. If your hot loop only needs a boolean like is_alive, an array-of-structs layout may still fetch a full object per element. His Monster example makes the tradeoff … https://reading-list.oddship.net/notes/2026-08-04-amp-orbs-agent-workspaces/ https://reading-list.oddship.net/notes/2026-08-04-amp-orbs-agent-workspaces/ Tue, 04 Aug 2026 22:36:00 +0530 Logged at IST: 2026-08-04 22:36 IST What it is: Thorsten Ball's Amp note on why Amp's orbs, remote sandboxed agent workspaces, changed how his team uses coding agents. Gist: Ball's argument is that the ingredients sound ordinary: secure sandbox, scale to zero, ephemeral workspaces, durable agent loops, web/phone/desktop control, previews, terminal, file editor, review panel, multiplayer, and automations. The hard part is conveying the felt change once those pieces remove local friction. The concrete shift is that he now spawns many more agents. A papercut can become a screenshot plus an agent … https://reading-list.oddship.net/notes/2026-08-04-pi-minimal-performant/ https://reading-list.oddship.net/notes/2026-08-04-pi-minimal-performant/ Tue, 04 Aug 2026 22:30:00 +0530 Logged at IST: 2026-08-04 22:30 IST What it is: Pi linking Earendil's essay on why Pi's minimal coding-agent harness design can be a performance and cost advantage. Gist: Earendil argues that Pi's small default surface is the point: four tools, plus system prompt and tool definitions under 1,000 tokens. The claim is that a harness should stay out of the model's way, preserve context discipline, and let users add workflow-specific complexity only when it earns its keep. The essay uses two external cases. In Databricks' internal benchmark on real work from a multi-million-line codebase, the same… https://reading-list.oddship.net/notes/2026-08-04-ai-aesthetic/ https://reading-list.oddship.net/notes/2026-08-04-ai-aesthetic/ Tue, 04 Aug 2026 13:24:00 +0530 Logged at IST: 2026-08-04 13:24 IST What it is: Jim Nielsen's short post on the emerging visual and interaction language around AI software. Gist: Nielsen frames AI as another computing zeitgeist that is producing its own design idioms. Some will disappear as fashion. Some may sink into the common language of software, the way the hamburger menu came from mobile constraints and then spread everywhere. The obvious symbols are sparkles and rainbow gradients, which now read as "AI" almost by default. The more interesting patterns are interaction details: streaming text as the native chat-interfac… https://reading-list.oddship.net/notes/2026-08-04-herdr-95-percent-less-cpu/ https://reading-list.oddship.net/notes/2026-08-04-herdr-95-percent-less-cpu/ Tue, 04 Aug 2026 11:57:00 +0530 Logged at IST: 2026-08-04 11:57 IST What it is: Can Celik's Herdr engineering post on cutting CPU use in multi-agent terminal sessions by avoiding frames that do not carry new information. Gist: The headline claim is not that Herdr's renderer became dramatically faster. It is that Herdr asks the renderer to do much less work. Across workloads dominated by unnecessary rendering, total CPU for the server plus attached clients fell by 89 to 95 percent. The three changes are nicely concrete. First, Herdr removed animated sidebar spinners for working agents. The static coloured state mark still tel… https://reading-list.oddship.net/notes/2026-08-04-llms-reward-expertise/ https://reading-list.oddship.net/notes/2026-08-04-llms-reward-expertise/ Tue, 04 Aug 2026 10:26:00 +0530 Logged at IST: 2026-08-04 10:26 IST What it is: Mario Zechner linking Sean Goedecke's essay on why LLMs reward domain expertise rather than making prompting skill-less. Gist: Goedecke's core claim is that the most important skill in prompting is expertise in the domain you are prompting for. LLMs let everyone produce passable work outside their specialty, but an expert can get much more from the same model because they know what to ask, what to ignore, where to push back, and what a good solution should feel like. His example is Terence Tao using ChatGPT on the Jacobian Conjecture. The visible… https://reading-list.oddship.net/notes/2026-08-03-yegge-agentic-harnesses-model-welfare/ https://reading-list.oddship.net/notes/2026-08-03-yegge-agentic-harnesses-model-welfare/ Mon, 03 Aug 2026 13:23:00 +0530 Logged at IST: 2026-08-03 13:23 IST What it is: Steve Yegge linking two long essays: Part 1 on agentic engineering harnesses, loops, graphs, Beads, Wheelhouse, CI/CD, code review, and the Wish Factory; Part 2 on his argument for model welfare as an engineering design constraint. Gist: The practical engineering claim is that serious agentic development stops looking like a smarter IDE and starts looking like a small city. Yegge's Wheelhouse setup for Wyvern has named crew agents that design work, fleet agents that implement it, role agents that operate parts of production, Beads as the work gra… https://reading-list.oddship.net/notes/2026-08-03-qwen3-8-max-frontend-code-arena/ https://reading-list.oddship.net/notes/2026-08-03-qwen3-8-max-frontend-code-arena/ Mon, 03 Aug 2026 12:04:00 +0530 Logged at IST: 2026-08-03 12:04 IST What it is: Arena.ai says Alibaba's Qwen3.8-Max landed at #4 on the Frontend Code Arena leaderboard, while Qwen's own launch post frames the model as a 2.4T-parameter, 95B-active model focused on coding, work, research, multimodal, and long-horizon tasks. Gist: The Arena result puts Qwen3.8-Max at 1,668 points, behind Claude Opus 5 Max at 1,705 and Kimi K3 Max at 1,676, roughly tied with Claude Opus 5 High at 1,669. Arena also says it ranks #2 in Consumer Product, #3 in Brand & Marketing, Reference-based Design, Gaming, and Content Creation Tools, #4 in … https://reading-list.oddship.net/notes/2026-08-03-devtools-must-be-open-source/ https://reading-list.oddship.net/notes/2026-08-03-devtools-must-be-open-source/ Mon, 03 Aug 2026 01:28:00 +0530 Logged at IST: 2026-08-03 01:28 IST What it is: David Crawshaw / exe.dev essay arguing that developer tools need to be open source because agent-driven personalization depends on editable source. Gist: Crawshaw’s core claim is that agents change the ROI of customizing software twice over: they make it cheap to start modifying a tool, and they can automate the ongoing work of rebasing those changes on upstream releases. In that world, source code becomes the extension system. Config files, plugin APIs, and vendor-provided hooks still help, but they are the old constrained surface; the powerful … https://reading-list.oddship.net/notes/2026-08-01-terminal-multiplexers-server-session-client-view/ https://reading-list.oddship.net/notes/2026-08-01-terminal-multiplexers-server-session-client-view/ Sat, 01 Aug 2026 13:45:00 +0530 Logged at IST: 2026-08-01 13:45 IST What it is: Peter Pistorius arguing that the current wave of new terminal multiplexers is about a real shift in how developers use computers, especially with long-running agents and shared remote compute. Gist: The core sentence is: “The server should own the session. The client should own the view.” Traditional multiplexers were designed around a human actively operating a terminal, with server-side windows, panes, focus, and layout. Peter argues that model breaks down when agents work for hours, multiple people or agents need to inspect the same work, and … https://reading-list.oddship.net/notes/2026-08-01-simon-willison-deepseek-v4-flash-0731/ https://reading-list.oddship.net/notes/2026-08-01-simon-willison-deepseek-v4-flash-0731/ Sat, 01 Aug 2026 12:48:00 +0530 Logged at IST: 2026-08-01 12:48 IST What it is: Simon Willison's link-blog note on deepseek-ai/DeepSeek-V4-Flash-0731, pointing to the Hugging Face release and Artificial Analysis' pricing/intelligence view. Gist: Simon highlights DeepSeek-V4-Flash-0731 as the latest V4-family release with “substantially enhanced agentic capabilities.” The model card says it is a 304B-parameter release, about 167GB on Hugging Face, that outperforms the V4-Pro preview on listed agent/code benchmarks despite a much smaller activated-parameter count. The important framing is value, not just model size. Simon note… https://reading-list.oddship.net/notes/2026-08-01-deepseek-v4-flash-high-frontend-code-arena/ https://reading-list.oddship.net/notes/2026-08-01-deepseek-v4-flash-high-frontend-code-arena/ Sat, 01 Aug 2026 12:32:00 +0530 Logged at IST: 2026-08-01 12:32 IST What it is: Arena.ai saying DeepSeek-V4-Flash-High has reshaped the Frontend Code Arena Pareto frontier with an arena score of 1586. Gist: The post's claim is not just that DeepSeek has another strong model. It is that DeepSeek-V4-Flash-High is unusually cheap for where it lands on the frontend-code price/performance curve. Arena lists it at $0.14/$0.28 per million tokens in the post, while the attached chart shows it around $0.25/M blended price, with a 1586 score. The chart places it on the Pareto frontier alongside much more expensive models: Claude Opus … https://reading-list.oddship.net/notes/2026-08-01-simon-willison-stateless-mcp/ https://reading-list.oddship.net/notes/2026-08-01-simon-willison-stateless-mcp/ Sat, 01 Aug 2026 11:41:00 +0530 Logged at IST: 2026-08-01 11:41 IST What it is: Simon Willison on why the 2026-07-28 stateless MCP specification has renewed his interest in MCP, with three projects he built around it: mcp-explorer, datasette-mcp, and llm-mcp-client. Gist: The technical hook is that stateless MCP collapses the older session flow into a single HTTP request. Instead of initializing a session, storing a Mcp-Session-Id, and then routing later tool calls against that state, a client can call a tool directly with an MCP-Protocol-Version header. That is cleaner for clients, simpler for servers, and a much better fit… https://reading-list.oddship.net/notes/2026-08-01-run-your-own-buzz-relay/ https://reading-list.oddship.net/notes/2026-08-01-run-your-own-buzz-relay/ Sat, 01 Aug 2026 11:38:00 +0530 Logged at IST: 2026-08-01 11:38 IST What it is: Block Engineering's practical guide to running a Buzz relay yourself, first on a laptop with Docker Compose and then on Railway or a VPS. Gist: This is the operational counterpart to the Buzz launch post. The relay is a single Rust binary that serves the WebSocket relay, REST API, and web UI, backed by Postgres, Redis, and S3-compatible object storage. The guide walks through the production Compose bundle, local startup, buzz-admin membership management, and joining the relay from Buzz Desktop. The important part is identity. A Buzz relay has its… https://reading-list.oddship.net/notes/2026-08-01-go-1-27-interactive-tour/ https://reading-list.oddship.net/notes/2026-08-01-go-1-27-interactive-tour/ Sat, 01 Aug 2026 11:32:00 +0530 Logged at IST: 2026-08-01 11:32 IST What it is: VictoriaMetrics' interactive tour of Go 1.27, turning the dry release notes into runnable examples across the language, runtime, standard library, and toolchain. Gist: The headline is generic methods. Methods can now declare their own type parameters, which makes operations like Box[T].Map[U] fit naturally on a type instead of living as package-level helpers. There is still an important boundary: interfaces cannot declare generic methods. The rest is a broad set of practical changes: promoted fields can be used directly as struct-literal keys, ge… https://reading-list.oddship.net/notes/2026-07-31-amazon-claude-cost-overruns/ https://reading-list.oddship.net/notes/2026-07-31-amazon-claude-cost-overruns/ Fri, 31 Jul 2026 10:11:00 +0530 Logged at IST: 2026-07-31 10:11 IST What it is: Tom's Hardware summarizing Financial Times reporting on internal Amazon AI usage metrics and Claude cost overruns. Gist: The article says Amazon internal reports found AI-agent projects blowing past budgets. The headline example is a failed Claude Sonnet deployment meant to match author details with Amazon listings: it reportedly cost $1.8 million, ran 860% over budget, and was only detected after about five months. Other cited overruns include $541,000 on a financial auditing tool and $134,000 on a logistics delivery-time project. Amazon's respo… https://reading-list.oddship.net/notes/2026-07-31-openai-gpt-5-6-pricing-cost-per-task/ https://reading-list.oddship.net/notes/2026-07-31-openai-gpt-5-6-pricing-cost-per-task/ Fri, 31 Jul 2026 10:05:00 +0530 Logged at IST: 2026-07-31 10:05 IST What it is: OpenAI announcing price and speed changes for GPT-5.6 Luna, Terra, and Sol across the API, Codex, and ChatGPT Work usage accounting. Gist: OpenAI says GPT-5.6 Luna pricing is dropping by 80%, GPT-5.6 Terra by 20%, and GPT-5.6 Sol is getting a faster API option. The attached chart frames Luna as the highest-scoring model on Artificial Analysis Intelligence Index v4.1 while being much cheaper per task than other frontier models in the chart. The product detail that matters is not just API sticker price: OpenAI says the lower Luna and Terra prices a… https://reading-list.oddship.net/notes/2026-07-29-ripgrep-kernel-bug-ai-debugging-filters/ https://reading-list.oddship.net/notes/2026-07-29-ripgrep-kernel-bug-ai-debugging-filters/ Wed, 29 Jul 2026 18:48:00 +0530 Logged at IST: 2026-07-29 18:48 IST What it is: A cluster of Daniel Franke and Perry Metzger posts around ripgrep issue #3494, grounded in Franke’s follow-up analysis repo rather than the X thread alone. Gist: The original bug looked like a weird ripgrep crash: the static x86_64-unknown-linux-musl build of ripgrep 15.2.0 could occasionally segfault during very large, highly concurrent directory searches. The issue report reproduced it with a ~20 GiB tree across about 1.8 million files, with crashes inside musl mallocng’s get_meta() path via calloc and opendir. Franke’s later analysis argues th… https://reading-list.oddship.net/notes/2026-07-29-mcp-2026-07-28-production-http-infrastructure/ https://reading-list.oddship.net/notes/2026-07-29-mcp-2026-07-28-production-http-infrastructure/ Wed, 29 Jul 2026 18:30:00 +0530 Logged at IST: 2026-07-29 18:30 IST What it is: Anthropic’s Claude product note, paired with the official MCP 2026-07-28 specification and release announcement. Gist: MCP 2026-07-28 is the release where MCP starts looking less like a demo-era integration protocol and more like ordinary production HTTP infrastructure. The core change is statelessness: the spec removes the initialize / initialized handshake and Mcp-Session-Id, makes each request self-contained with protocol version, client identity, and capabilities in _meta, and adds optional server/discover for clients that want server capabil… https://reading-list.oddship.net/notes/2026-07-29-camelai-agent-durable-objects/ https://reading-list.oddship.net/notes/2026-07-29-camelai-agent-durable-objects/ Wed, 29 Jul 2026 01:57:00 +0530 Logged at IST: 2026-07-29 01:57 IST What it is: Miguel Salinas explains how camelAI rewrote its coding agent stack to stop running each user on an always-on virtual machine. Gist: The old camelAI setup used the Claude Code harness and a self-built VM/container service. That worked, but always-on machines plus attached disk were too expensive for the scale they wanted. The rewrite moved the “brain” of the agent into a Cloudflare Durable Object first, then removed the VM-backed project runtime entirely. The current stack stores each project filesystem in Durable Object SQLite, with larger files … https://reading-list.oddship.net/notes/2026-07-28-tsgo-as-go-tool/ https://reading-list.oddship.net/notes/2026-07-28-tsgo-as-go-tool/ Tue, 28 Jul 2026 20:59:00 +0530 Logged at IST: 2026-07-28 20:59 IST What it is: Ben Visness points out that Microsoft’s Go-native TypeScript port can be added to a go.mod file and run with go tool tsgo. Gist: The screenshot shows this Go tool directive: tool github.com/microsoft/typescript-go/cmd/tsgo The practical point is that a Go project can pin and invoke tsgo through Go’s own tool dependency workflow, without installing npm just to run TypeScript. The microsoft/typescript-go repository describes itself as the staging repo for the native port of TypeScript, and its own go.mod is already structured as a normal Go module… https://reading-list.oddship.net/notes/2026-07-28-cursor-agent-swarm-model-economics/ https://reading-list.oddship.net/notes/2026-07-28-cursor-agent-swarm-model-economics/ Tue, 28 Jul 2026 19:37:00 +0530 Logged at IST: 2026-07-28 19:37 IST What it is: Cursor research post by Wilson Lin on the next version of their agent swarm system, tested by asking agents to implement SQLite in Rust from the 835-page manual. Gist: Cursor’s claim is that agent swarms scale because they split context, not just because they run in parallel. Planner agents use frontier models to decompose a goal and make design decisions. Worker agents use faster and cheaper models to execute narrow leaves without carrying the whole task tree in context. The SQLite experiment compares old and new swarms on the same task and time… https://reading-list.oddship.net/notes/2026-07-28-ghostty-alacritty-terminal-grid-memory/ https://reading-list.oddship.net/notes/2026-07-28-ghostty-alacritty-terminal-grid-memory/ Tue, 28 Jul 2026 13:27:00 +0530 Logged at IST: 2026-07-28 13:27 IST What it is: Mitchell Hashimoto explains why Ghostty/libghostty's terminal grid is much more memory-efficient than Alacritty's terminal crate in his benchmark. Gist: The core comparison is simple: Alacritty stores 32 bytes of metadata per row and 24 bytes per cell, while Ghostty represents each row and each cell in 8 bytes. Mitchell says around 95% of the uncompressed memory-usage difference comes from a few data-structure choices. Ghostty does not store the full style beside every cell. It stores a 16-bit style ID and dedupes styles in a look-aside, referenc… https://reading-list.oddship.net/notes/2026-07-27-deer-workflow-dynamic-workflow-runtime/ https://reading-list.oddship.net/notes/2026-07-27-deer-workflow-dynamic-workflow-runtime/ Mon, 27 Jul 2026 04:40:00 +0530 Logged at IST: 2026-07-27 04:40 IST What it is: Deer Workflow is an open-source Dynamic Workflow runtime for AI agents, published as @deerwork-ai/deer-workflow with a deer-workflow CLI. Gist: The project argues for a middle ground between traditional workflows and autonomous agents: keep orchestration, order, concurrency, phases, and schemas in ordinary TypeScript, and call an agent only where judgment is needed. Its visible slogan captures the split: “Let code drive the flow. Agents handle judgment.” The implementation matches that framing. Workflow modules export a handler and optional stati… https://reading-list.oddship.net/notes/2026-07-27-auto-trees-prime-vs-pi-weaver/ https://reading-list.oddship.net/notes/2026-07-27-auto-trees-prime-vs-pi-weaver/ Mon, 27 Jul 2026 04:37:00 +0530 Logged at IST: 2026-07-27 04:37 IST What it is: A comparison between Auto Trees' new /prime <scope> command for Pi and pi-weaver, using the public oddship/bosun/packages/pi-weaver implementation Rohan pointed me to. Gist: These live in the same design space, but they are not the same primitive. Auto Trees asks, “Where should the next clean working context start?” Pi-weaver asks, “When should this failed branch stop existing?” Auto Trees is a human-triggered session hygiene tool. /prime <scope> sends Pi a scoped orientation prompt, asks it to map only relevant code, avoid dependency… https://reading-list.oddship.net/notes/2026-07-27-tigerbeetle-object-storage-vs-simples3/ https://reading-list.oddship.net/notes/2026-07-27-tigerbeetle-object-storage-vs-simples3/ Mon, 27 Jul 2026 04:31:00 +0530 Logged at IST: 2026-07-27 04:31 IST What it is: TigerBeetle shared notes around its custom object-storage clients, and I compared that design direction with rhnvrm/simples3. Gist: The shared instinct is anti-SDK: avoid a large vendor dependency tree, understand the REST/signing surface directly, and keep the client small enough to reason about. But the two implementations optimize for different worlds. simples3 is a pragmatic Go S3 REST client. It is stdlib-only, supports SigV4, custom endpoints, bucket/object/list/multipart/tagging/versioning/SSE/IAM features, and exposes a straightforward bl… https://reading-list.oddship.net/notes/2026-07-26-why-software-factories-fail-part-two/ https://reading-list.oddship.net/notes/2026-07-26-why-software-factories-fail-part-two/ Sun, 26 Jul 2026 03:34:00 +0530 Logged at IST: 2026-07-26 03:34 IST What it is: Dex Horthy’s part two of “Why Software Factories Fail,” published as an X Article. Gist: Dex’s follow-up says the answer to unreliable lights-out software factories is not more faith in model judges. For now, the judge is still the human, so the right move is to put review and planning back into the loop earlier. He proposes four leverage points: product requirements, system architecture, program design, and vertical slices. Product review pins down the user pain and success condition. Architecture aligns services, endpoints, schemas, queues, and… https://reading-list.oddship.net/notes/2026-07-26-antirez-being-linus-torvalds/ https://reading-list.oddship.net/notes/2026-07-26-antirez-being-linus-torvalds/ Sun, 26 Jul 2026 03:32:00 +0530 Logged at IST: 2026-07-26 03:32 IST What it is: Antirez post adapting the transcript of his video, “Being Linus Torvalds,” into English. Gist: Antirez argues that Linus Torvalds’s real genius was not just writing the first Linux kernel. Other very strong programmers could have built a small working kernel. What made Linus singular was that he stepped back early from writing most code and became the person holding the kernel’s design direction, taste, and hierarchy of decisions. He maps that role onto AI-assisted programming. Expert programmers using agents well are not simply asking for code a… https://reading-list.oddship.net/notes/2026-07-25-letta-trajectory-agent-experience-data/ https://reading-list.oddship.net/notes/2026-07-25-letta-trajectory-agent-experience-data/ Sat, 25 Jul 2026 12:49:00 +0530 Logged at IST: 2026-07-25 12:49 IST What it is: Letta blog post introducing @letta-ai/trajectory, a normalized format and package for coding-agent session data. Gist: Letta argues that agents can learn from past sessions across Claude Code, Codex, Letta Code, and other harnesses only if the experience is normalized into a token-efficient, agent-readable format. trajectory represents user messages, assistant messages, reasoning, tool calls, tool results, and harness metadata in one standard record schema. The key design choice is to optimize for agents reading past sessions, not full-fidelity r… https://reading-list.oddship.net/notes/2026-07-25-context-engineering-claude-5/ https://reading-list.oddship.net/notes/2026-07-25-context-engineering-claude-5/ Sat, 25 Jul 2026 02:24:00 +0530 Logged at IST: 2026-07-25 02:24 IST What it is: Thariq / Claude Code X post with an embedded X Article, “The new rules of context engineering for Claude 5 models.” Gist: Claude Code removed more than 80% of its system prompt for Claude Opus 5 and Fable 5 with no measurable loss on coding evaluations. The article’s point is that stronger models need less rule-heavy scaffolding. Older context-engineering habits can now overconstrain the model, create contradictory instructions, and make Claude spend effort reconciling durable context instead of judging the user’s current intent. The new guidance… https://reading-list.oddship.net/notes/2026-07-24-observing-go-green-tea-gc/ https://reading-list.oddship.net/notes/2026-07-24-observing-go-green-tea-gc/ Fri, 24 Jul 2026 23:44:00 +0530 Logged at IST: 2026-07-24 23:44 IST What it is: Phil Eaton / The Consensus deep dive on observing Go’s Green Tea garbage collector and Go’s non-moving heap behavior. Gist: The article starts by making Go’s heap layout visible. Small, medium, and large objects are allocated into size-segregated spans, so randomly allocated objects end up grouped by size. C# is used as a contrast: its objects are not arranged the same way, and later examples show its compacting collector moving survivors together after cleanup. The Green Tea part is about cache locality. Go 1.26’s default collector scans spans a… https://reading-list.oddship.net/notes/2026-07-24-claude-opus-5/ https://reading-list.oddship.net/notes/2026-07-24-claude-opus-5/ Fri, 24 Jul 2026 22:52:00 +0530 Logged at IST: 2026-07-24 22:52 IST What it is: Anthropic product announcement for Claude Opus 5, positioned as a more efficient daily-use frontier model below Fable 5 but ahead on several coding and knowledge-work evaluations. Gist: Anthropic says Opus 5 keeps the same base pricing as Opus 4.8 while improving performance across coding, automation, computer use, knowledge work, visual outputs, and life-sciences tasks. A big part of the pitch is cost-aware effort control: customers can trade off intelligence, speed, and token use, with Opus 5 reportedly approaching Fable 5 on some coding tasks … https://reading-list.oddship.net/notes/2026-07-24-armada-encrypted-discord-alternative/ https://reading-list.oddship.net/notes/2026-07-24-armada-encrypted-discord-alternative/ Fri, 24 Jul 2026 18:10:00 +0530 Logged at IST: 2026-07-24 18:10 IST What it is: Soapbox launch/explainer for Armada, an open-source, end-to-end encrypted Discord-style community chat app built on Nostr and the Concord protocol. Gist: Armada is pitched as an answer to several Discord pain points: centralized bans, unencrypted text history, ID and face-scan verification pressure, data breaches, ads, and IPO-driven monetization. Instead of a company-issued account, identity is a Nostr key pair. Communities are end-to-end encrypted, publish encrypted messages to multiple independent relays, and can keep running if a relay disapp… https://reading-list.oddship.net/notes/2026-07-24-anthropic-ai-native-sdlc-security/ https://reading-list.oddship.net/notes/2026-07-24-anthropic-ai-native-sdlc-security/ Fri, 24 Jul 2026 17:50:00 +0530 Logged at IST: 2026-07-24 17:50 IST What it is: Anthropic post by Deputy CISO Jason Clinton on securing an AI-native software development lifecycle where Claude authors about 80% of merged code. Gist: Anthropic says its code volume and deployment velocity have scaled sharply, with engineers shipping 8x as much code per quarter as before and Claude authoring most merged code. The security response is not a single giant AI reviewer. It is a layered SDLC: project security reviews connected to organizational context, security guidance encoded into CLAUDE.md and skills, remote coding VMs with tight… https://reading-list.oddship.net/notes/2026-07-24-why-software-factories-fail/ https://reading-list.oddship.net/notes/2026-07-24-why-software-factories-fail/ Fri, 24 Jul 2026 17:32:00 +0530 Logged at IST: 2026-07-24 17:32 IST What it is: Dex Horthy / HumanLayer essay arguing against lights-off coding-agent software factories. Gist: The post argues that harness engineering, loop engineering, and benchmark-driven agent gains make agents much faster at producing code, but they do not solve long-term maintainability. Current coding-agent training and evaluation reward short-horizon success, mostly “did the tests pass?”, while bad design shows up weeks or months later as shotgun surgery, slow review, and fragile systems. The practical answer is not to abandon agents, but to turn the l… https://reading-list.oddship.net/notes/2026-07-23-claude-is-not-a-compiler/ https://reading-list.oddship.net/notes/2026-07-23-claude-is-not-a-compiler/ Thu, 23 Jul 2026 08:28:00 +0530 Logged at IST: 2026-07-23 08:28 IST What it is: exe.dev essay arguing that Claude and coding agents are not compilers, but vertically integrated engineering resources. Gist: The post rejects the idea that Claude is merely a natural-language-to-code compiler. A compiler hides one lower-level decision layer; a coding agent can move across strategy, product, architecture, implementation, tests, adversarial review, and operational failure modes. The concrete exe.dev DNS example shows agents producing divergent full-system implementations, surfacing cross-layer questions, and helping the author bui… https://reading-list.oddship.net/notes/2026-07-23-beej-ai-making/ https://reading-list.oddship.net/notes/2026-07-23-beej-ai-making/ Thu, 23 Jul 2026 08:04:00 +0530 Logged at IST: 2026-07-23 08:04 IST What it is: Beej’s essay on AI, authorship, and the felt difference between making something and having it made for you. Gist: Beej grants the familiar AI-dev tradeoffs around craft, low-level problem-solving, and finishing back-burnered projects, but says his sharper loss is the loss of making. He is uncomfortable claiming credit for AI-generated prose, art, code, or contractor-built work because initiating and managing creation feels different from making it himself. Prompting has skill and judgment, but for him it is closer to asking someone else to make … https://reading-list.oddship.net/notes/2026-07-23-everyone-should-know-simd/ https://reading-list.oddship.net/notes/2026-07-23-everyone-should-know-simd/ Thu, 23 Jul 2026 07:25:00 +0530 Logged at IST: 2026-07-23 07:25 IST What it is: Mitchell Hashimoto linking to his practical SIMD introduction, “Everyone Should Know SIMD,” using a Ghostty loop as the worked example. Gist: Hashimoto argues that everyday SIMD should not feel scary: the common “process N values at a time” case follows a repeatable five-step shape: broadcast constants, loop one vector-width chunk at a time, do the parallel operation, reduce or store the vector result, then finish with the scalar tail. His Ghostty example turns a scalar codepoint scan into generic Zig vector code, yielding up to 4x, 8x, or 16x la… https://reading-list.oddship.net/notes/2026-07-22-antirez-software-distribution-with-coding-agents/ https://reading-list.oddship.net/notes/2026-07-22-antirez-software-distribution-with-coding-agents/ Wed, 22 Jul 2026 21:35:00 +0530 Logged at IST: 2026-07-22 21:35 IST What it is: antirez linking to his blog post “Not just development, distribution of software may change as well.” Gist: antirez argues that AI changes not only how software is developed, but how it can be distributed and used. If users have coding agents, a repository can be more than a polished stable artifact: it can be a template, a set of rails, and a collection of experimental branches that users adapt to their hardware, models, and requirements. His Redis and DwarfStar examples point toward more fluid release practices where 95%-ready branches, agent-r… https://reading-list.oddship.net/notes/2026-07-22-github-stargazer-api-restriction/ https://reading-list.oddship.net/notes/2026-07-22-github-stargazer-api-restriction/ Wed, 22 Jul 2026 16:55:00 +0530 Logged at IST: 2026-07-22 16:55 IST What it is: Star History’s note on GitHub restricting access to the stargazers API. Gist: GitHub is limiting the endpoint that reveals who starred a repository and when to repository admins and collaborators. Star History depends on exactly that data to reconstruct growth curves, so charts for repositories you do not own or collaborate on are broken, and README-embedded live charts are affected because Star History’s servers are not collaborators on most repos. Owner and collaborator views still work with a readable fine-grained token; old no-scope tokens no… https://reading-list.oddship.net/notes/2026-07-22-coding-agent-microvms-with-nix/ https://reading-list.oddship.net/notes/2026-07-22-coding-agent-microvms-with-nix/ Wed, 22 Jul 2026 16:10:00 +0530 Logged at IST: 2026-07-22 16:10 IST What it is: Michael Stapelberg’s NixOS and microvm.nix setup for running coding agents inside ephemeral MicroVMs. Gist: Stapelberg wants coding agents to run without per-command review while keeping them away from personal files and making compromise disposable. His setup uses NixOS, microvm.nix, a NATed bridge, shared project workspaces, shared Claude credentials, cloud-hypervisor, and home-manager so each project gets a reproducible ephemeral VM. He also shows a Claude Skill that creates new project MicroVM definitions, picks free IPs, clones repositories,… https://reading-list.oddship.net/notes/2026-07-22-history-of-ides-at-google/ https://reading-list.oddship.net/notes/2026-07-22-history-of-ides-at-google/ Wed, 22 Jul 2026 16:07:00 +0530 Logged at IST: 2026-07-22 16:07 IST What it is: Laurent Le Brun’s history of IDEs at Google, focused on google3, Cider, and Cider V. Gist: Le Brun traces Google’s shift from editor pluralism to a de facto standard internal IDE. The core constraint was scale: local IDE assumptions broke against a billion-file monorepo, shared build, search, review tooling, and a language graph that had to match each developer’s sync state. Cider solved much of this with a web editor backed by centralized indexing, then Cider V adopted VS Code as the frontend so Google-specific backend investment could combine w… https://reading-list.oddship.net/notes/2026-07-22-freerange-verifiable-ui-ranges/ https://reading-list.oddship.net/notes/2026-07-22-freerange-verifiable-ui-ranges/ Wed, 22 Jul 2026 14:07:00 +0530 Logged at IST: 2026-07-22 14:07 IST What it is: Cheng Lou announcing Freerange, a zero-API static analysis tool for verifiable user interfaces. Gist: Freerange aims to turn vibe coding toward proof engineering by automatically deducing numerical ranges in code. Lou says it can statically prove that TypeScript layouts obey specified sizing, avoid NaN and Infinity, and keep array indices within bounds, with no browser or runtime execution. He also positions the approach as RL-friendly for ML workflows. Newsletter angle: Useful developer-tools item because it makes UI correctness concrete: static… https://reading-list.oddship.net/notes/2026-07-22-buzz-channel-driven-agent-workspace/ https://reading-list.oddship.net/notes/2026-07-22-buzz-channel-driven-agent-workspace/ Wed, 22 Jul 2026 14:06:00 +0530 Logged at IST: 2026-07-22 14:06 IST Update, 2026-07-27: Matt Rice shared a roundup of reported Buzz use cases since launch: an 11-agent company migration with orchestration in Buzz; a delegate-only "Chief" agent that staffs channels with specialists; a WordPress content factory that writes, publishes, and verifies pages; a Linear plus E2B bridge where assigning an issue wakes a manager agent and runs work in a fresh sandbox; shared local compute endpoints for communities; Hermes/GB10 pooled inference organizing; and voice-driven setup through Codex. Treat these as reported examples rather than… https://reading-list.oddship.net/notes/2026-07-22-quality-software/ https://reading-list.oddship.net/notes/2026-07-22-quality-software/ Wed, 22 Jul 2026 00:37:00 +0530 Logged at IST: 2026-07-22 00:37 IST What it is: Mitchell Hashimoto recommending Alasdair Monk’s X article Quality Software. Gist: Monk argues that AI lowering the barrier to software creation also lowers quality unless people aim at quality deliberately. His definition is intentionally plain: quality software does not break, does not demand attention, knows its limits, and fixes fast. The AI-specific point is that “slop” is not new, but AI produces it faster, and the rush to “be agentic” can make companies forget why users chose the software in the first place. His sharpest boundary is that ag… https://reading-list.oddship.net/notes/2026-07-21-martin-fowler-fragments-july-21/ https://reading-list.oddship.net/notes/2026-07-21-martin-fowler-fragments-july-21/ Tue, 21 Jul 2026 21:06:00 +0530 Logged at IST: 2026-07-21 21:06 IST What it is: Martin Fowler’s July 21 fragment wrapping up notes from the second Future of Software Development Retreat, plus related fragments on legal education, DSLs, and LLM-speak. Gist: Fowler’s retreat notes line up with the recent factory and harness theme: code generation is no longer the bottleneck; verification is. The Thoughtworks report headlines harness engineering as an ownable discipline, flags an apprenticeship crisis, warns that executive expectations are outrunning engineering risk judgment, and sees legacy modernization as the most defensibl… https://reading-list.oddship.net/notes/2026-07-21-one-document-two-hands/ https://reading-list.oddship.net/notes/2026-07-21-one-document-two-hands/ Tue, 21 Jul 2026 19:31:00 +0530 Logged at IST: 2026-07-21 19:31 IST What it is: Sunil Pai sharing one document, two hands, the written version of his Local-First Conf talk on embedding agent harnesses into ordinary apps. Gist: Pai argues that coding agents feel powerful because developers gave them a real workshop: repositories, shells, editors, tests, tools, and runnable feedback loops. The broader product lesson is not to put chat in front of every app, but to let the agent work beside the user on the same document. In his Pizzo demo, the user edits a song through normal controls while the agent calls the same deterministi… https://reading-list.oddship.net/notes/2026-07-21-software-factories-light-and-dark/ https://reading-list.oddship.net/notes/2026-07-21-software-factories-light-and-dark/ Tue, 21 Jul 2026 19:19:00 +0530 Logged at IST: 2026-07-21 19:19 IST What it is: Addy Osmani’s X article Software Factories, Light and Dark, riffing on Dex Horthy’s talk about why software factories fail. Gist: Osmani argues that agentic software factories are not bigger agents, but many harnessed loops fed by queues and drained through review gates. The dark version removes human reading from the floor and lets agents scope, build, verify, and ship with only machine checks. That creates apparent throughput while accumulating comprehension debt: code expands faster than any human understands it, with tests green until the lat… https://reading-list.oddship.net/notes/2026-07-19-pattern-matching-strings-without-decompression/ https://reading-list.oddship.net/notes/2026-07-19-pattern-matching-strings-without-decompression/ Sun, 19 Jul 2026 23:32:00 +0530 Logged at IST: 2026-07-19 23:32 IST What it is: Spiral/Vortex deep dive on evaluating SQL LIKE predicates directly over FSST-compressed strings. Gist: The post explains how Vortex can match string patterns in compressed code space instead of decompressing values first. It builds deterministic finite automata over FSST symbol codes, then uses a SIMD Teddy prefilter to cheaply identify candidate positions before running the verifier. On selective patterns like ClickBench Q20’s %google%, the prefilter cuts candidate stops from about ten million to 44,158 and confirms 646 true matches, producing r… https://reading-list.oddship.net/notes/2026-07-19-the-zero-cost-fallacy-open-source-software-in-the-agentic-era/ https://reading-list.oddship.net/notes/2026-07-19-the-zero-cost-fallacy-open-source-software-in-the-agentic-era/ Sun, 19 Jul 2026 16:28:00 +0530 Logged at IST: 2026-07-19 16:28 IST What it is: Bilgin Ibryam sharing Thoughtworks’ The zero-cost fallacy: Open source software in the agentic era. Gist: The article argues that the agentic era intensifies long-running open-source sustainability problems. It separates zero marginal distribution cost from real maintenance labor, then adds two AI-era pressures: low-effort generated pull requests that turn maintainers into unpaid reviewers, and a degraded trust landscape where stars, recency, and apparent activity are easier to manipulate. It also frames permissive licensing as part of an extract… https://reading-list.oddship.net/notes/2026-07-19-ktransformers-and-heterogeneous-moe-inference/ https://reading-list.oddship.net/notes/2026-07-19-ktransformers-and-heterogeneous-moe-inference/ Sun, 19 Jul 2026 02:31:00 +0530 Logged at IST: 2026-07-19 02:31 IST What it is: X post pointing to kvcache-ai/ktransformers, a Tsinghua MADSys Lab project for CPU-GPU heterogeneous inference and fine-tuning of large MoE models. Gist: The viral framing is a little breathless, but the underlying project is real and interesting: KTransformers is about heterogeneous CPU/GPU execution for large MoE models, keeping hot experts on GPU and offloading colder expert work to CPU/DRAM so very large models can be explored on commodity-ish hardware. The repo’s own docs claim DeepSeek-V3/R1 support on 24GB VRAM with long-context paths, 3x … https://reading-list.oddship.net/notes/2026-07-16-shard-your-locks/ https://reading-list.oddship.net/notes/2026-07-16-shard-your-locks/ Thu, 16 Jul 2026 23:42:00 +0530 Logged at IST: 2026-07-16 23:42 IST What it is: Misha Strebkov benchmarking six Go in-memory cache designs under different read/write mixes and core counts. Gist: The sharp result is that the boring “just use RWMutex for read-heavy caches” instinct does not hold up well under contention. A 256-shard striped map was the best all-around design, scaling up to about 8x faster than a single mutex at 8 cores, while RWMutex plateaued early and could even lose to a plain mutex on writes. The broader lesson is to reason about contention topology, cache-line movement, and workload skew, not just read/wr… https://reading-list.oddship.net/notes/2026-07-16-linus-torvalds-on-linux-and-ai-tools/ https://reading-list.oddship.net/notes/2026-07-16-linus-torvalds-on-linux-and-ai-tools/ Thu, 16 Jul 2026 23:37:00 +0530 Logged at IST: 2026-07-16 23:37 IST What it is: Linus Torvalds pushing back on anti-AI sentiment in kernel development and saying Linux is not an anti-AI project. Gist: Linus’s operative distinction is not "everyone must use AI," but that the kernel should evaluate tools on technical merit, not ideological discomfort. He argues AI is now clearly useful, maintainers should focus on making LLM tools reduce pain rather than banning them, and people who object to others using AI can fork or walk away. The broader point is that the kernel project is about better technology, not being a social-warri… https://reading-list.oddship.net/notes/2026-07-16-boris-cherny-on-domain-knowledge-as-infrastructure/ https://reading-list.oddship.net/notes/2026-07-16-boris-cherny-on-domain-knowledge-as-infrastructure/ Thu, 16 Jul 2026 01:33:00 +0530 Logged at IST: 2026-07-16 01:33 IST What it is: Boris Cherny arguing that agent-era engineering leverage still comes from automation, but now automation also includes encoded domain knowledge like CLAUDE.md, review rules, skills, and docs Gist: The core claim is that the old highest-leverage engineering move, turning recurring work into infrastructure, matters even more with agents. Better lint rules, CI steps, tests, routines, and DevX speed up both humans and agent swarms. More importantly, domain knowledge that used to live in people’s heads now needs to be encoded as machine-usable infrast… https://reading-list.oddship.net/notes/2026-07-15-fable-5-is-insane-i-vibe-coded-terminator-vision/ https://reading-list.oddship.net/notes/2026-07-15-fable-5-is-insane-i-vibe-coded-terminator-vision/ Wed, 15 Jul 2026 16:58:00 +0530 Logged at IST: 2026-07-15 16:58 IST What it is: Bilawal Sidhu video titled “Fable 5 Is Insane. I Vibe Coded Terminator Vision.” Gist: From the visible YouTube description and page metadata, this is a build/demo video about creating a browser-based range-analysis system from ordinary 2D video sources like Meta Ray-Bans, iPhones, and GoPros, then reconstructing shots, hits/misses, targets, and a replayable 3D "god’s eye" view of the session. Newsletter angle: Interesting computer-vision / spatial-reconstruction demo that fits the broader pattern of fast prototyping with modern models plus commod… https://reading-list.oddship.net/notes/2026-07-15-the-memory-heist/ https://reading-list.oddship.net/notes/2026-07-15-the-memory-heist/ Wed, 15 Jul 2026 13:30:00 +0530 Logged at IST: 2026-07-15 13:30 IST What it is: Ayush Paul’s writeup on prompt-injecting Claude’s memory and browsing system into exfiltrating personal data Gist: The attack chain was not about breaking the memory store directly, but about combining long-lived personal memory with a browsing agent that could be socially engineered into leaking data through link-by-link URL navigation. The important point is that once an assistant can search history, infer missing details, and autonomously browse attacker-controlled pages, "read-only" web access can still become an exfiltration channel. Newslet… https://reading-list.oddship.net/notes/2026-07-15-how-razorpay-refreshes-its-data-warehouse-10x-faster/ https://reading-list.oddship.net/notes/2026-07-15-how-razorpay-refreshes-its-data-warehouse-10x-faster/ Wed, 15 Jul 2026 10:45:00 +0530 Logged at IST: 2026-07-15 10:45 IST What it is: Piyush Goel sharing Razorpay Engineering’s writeup on refreshing warehouse facts 10x faster with graphs and indexes Gist: Razorpay moved from expensive full-refresh fact generation toward incremental fact maintenance by treating each denormalized fact as a dependency graph. They pair change-driven processing with secondary indexes on the lake, graph traversal to discover affected ancestors and descendants, and selective runtime joins for high-cardinality dimensions. The result is much faster warehouse refreshes with lower compute cost, restored h… https://reading-list.oddship.net/notes/2026-07-14-dsls-enable-reliable-use-of-llms/ https://reading-list.oddship.net/notes/2026-07-14-dsls-enable-reliable-use-of-llms/ Tue, 14 Jul 2026 20:54:00 +0530 Logged at IST: 2026-07-14 20:54 IST What it is: Martin Fowler sharing Unmesh Joshi’s article on DSLs and LLM reliability Gist: The article’s core claim is that LLMs become much more reliable when they are constrained by domain abstractions and DSLs instead of being asked to directly generate unconstrained general-purpose code. The deeper point is that DSLs do double duty: they help teams discover and stabilize a semantic model during design, and then they become a natural-language target that LLMs can generate against, validate, and repair with much tighter feedback loops. Newsletter angle: St… https://reading-list.oddship.net/notes/2026-07-14-dave-winer-introduces-rss-chat/ https://reading-list.oddship.net/notes/2026-07-14-dave-winer-introduces-rss-chat/ Tue, 14 Jul 2026 09:44:00 +0530 Logged at IST: 2026-07-14 09:44 IST What it is: Dave Winer introducing rss.chat and arguing for RSS as a social-network substrate Gist: Winer presents rss.chat as a small-community social system built from old web primitives: RSS 2.0, OPML, Markdown, SQL, WebSocket, and rssCloud. The pitch is that social publishing and reply structures do not need heavyweight new protocols if interoperable feeds, shared formats, and replaceable components are treated as the core product. Newsletter angle: Good example of AI-assisted software being used to revive old-web interoperability ideas, with a strong sm… https://reading-list.oddship.net/notes/2026-07-14-gergely-orosz-on-trust-burn-from-grok-cli-privacy-concerns/ https://reading-list.oddship.net/notes/2026-07-14-gergely-orosz-on-trust-burn-from-grok-cli-privacy-concerns/ Tue, 14 Jul 2026 08:14:00 +0530 Logged at IST: 2026-07-14 08:14 IST What it is: Gergely Orosz calling out reports that Grok CLI uploaded codebases without users knowingly consenting, quoting SpaceXAI’s privacy response Gist: The important issue here is not just data retention policy wording but trust boundary failure. Orosz’s point is that if developers believe a local coding tool silently uploaded proprietary code without clear consent, the damage is immediate and reputational, even if the vendor later points to settings like zero data retention or a /privacy command. Newsletter angle: Strong example of how AI devtools live… https://reading-list.oddship.net/notes/2026-07-14-using-uvx-in-github-actions-in-a-cache-friendly-way/ https://reading-list.oddship.net/notes/2026-07-14-using-uvx-in-github-actions-in-a-cache-friendly-way/ Tue, 14 Jul 2026 08:12:00 +0530 Logged at IST: 2026-07-14 08:12 IST What it is: Simon Willison linking to his TIL on running uvx in GitHub Actions without re-downloading the package every run Gist: The useful trick is to pin UV_EXCLUDE_NEWER to a date, use that same date in the GitHub Actions cache key, and set UV_OFFLINE=1 on cache hits. That gives a lightweight, file-free way to make uvx tool-name workflows cacheable, reproducible enough, and intentionally bustable by changing one date. Newsletter angle: Nice practical CI pattern for teams using uvx as disposable tooling glue, especially because it avoids adding fake depen… https://reading-list.oddship.net/notes/2026-07-13-control-the-ideas-not-the-code/ https://reading-list.oddship.net/notes/2026-07-13-control-the-ideas-not-the-code/ Mon, 13 Jul 2026 19:31:00 +0530 Logged at IST: 2026-07-13 19:31 IST What it is: X post by antirez linking his blog post "Control the ideas, not the code" Gist: antirez extends the earlier X-thread argument into a full workflow claim: if you own the ideas, design, testing, and QA of a system, then line-by-line review of generated code is increasingly the wrong bottleneck. He argues that models are already better at many local code checks than humans, and that the higher-leverage work is controlling the mental model, writing human-readable design docs, and spending time on quality and new ideas instead of staring at implementa… https://reading-list.oddship.net/notes/2026-07-13-mario-zechner-on-types-interfaces-and-reading-generated-code/ https://reading-list.oddship.net/notes/2026-07-13-mario-zechner-on-types-interfaces-and-reading-generated-code/ Mon, 13 Jul 2026 12:19:00 +0530 Logged at IST: 2026-07-13 12:19 IST What it is: X post by Mario Zechner adding nuance to antirez’s AI-code ownership point Gist: Zechner’s point is narrower and more practical than a general plea for control: if you control the types and interfaces, the rest often falls into place well enough. But current models still love to introduce bad abstractions that work against those boundaries, so in practice you sometimes have to read generated code and beat it back into submission instead of letting it stomp over the structure you intended. Newsletter angle: Useful follow-on to the "own the mental … https://reading-list.oddship.net/notes/2026-07-13-antirez-on-owning-the-mental-model-in-ai-coded-systems/ https://reading-list.oddship.net/notes/2026-07-13-antirez-on-owning-the-mental-model-in-ai-coded-systems/ Mon, 13 Jul 2026 07:49:00 +0530 Logged at IST: 2026-07-13 07:49 IST What it is: X post by antirez on the "don't look at the code" debate in AI-coded systems Gist: He distinguishes between two very different ways an AI-coded codebase can come into existence: one where the human still controls the main ideas and keeps a coherent mental model of the system, and one where the human brute-forces prompts until something works. The point is that these may look similar from the outside but carry very different implications for understanding, maintainability, and trust. Newsletter angle: Sharp framing for a real fault line in AI-assi… https://reading-list.oddship.net/notes/2026-07-13-old-and-new-apps-via-modern-coding-agents/ https://reading-list.oddship.net/notes/2026-07-13-old-and-new-apps-via-modern-coding-agents/ Mon, 13 Jul 2026 07:37:00 +0530 Logged at IST: 2026-07-13 07:37 IST What it is: X post by Mario Zechner recommending Terry Tao’s post "Old and new apps, via modern coding agents" Gist: Tao describes using modern coding agents to port his old Java applets to JavaScript and revive them quickly, with surprisingly low bug overhead, then goes further and uses the same workflow to build new math visualization tools he had wanted for decades. The interesting point is not just vibe coding as novelty, but coding agents as leverage for software archaeology, maintenance, and low-risk supplementary tooling. Newsletter angle: Strong data… https://reading-list.oddship.net/notes/2026-07-12-vim-of-coding-agents/ https://reading-list.oddship.net/notes/2026-07-12-vim-of-coding-agents/ Sun, 12 Jul 2026 00:00:00 +0000 Logged at IST: 2026-07-12 13:18 IST What it is: X post by dogfiles linking the blog post "Vim of Coding Agents" Gist: Frames Pi as the Neovim of coding agents: a minimal, hackable foundation that adapts to your workflow instead of forcing you into an opinionated all-in-one agent product. The writeup argues that the real value is not just using Pi as shipped, but treating it as a customizable harness where you can build your own tools, TUI tweaks, prompts, and extensions. Newsletter angle: Good articulation of the coding-agent split between turnkey products and configurable harnesses, especiall… https://reading-list.oddship.net/notes/2026-07-11-101-saying-a-go-proposal-was-formally-rejected/ https://reading-list.oddship.net/notes/2026-07-11-101-saying-a-go-proposal-was-formally-rejected/ Sat, 11 Jul 2026 00:00:00 +0000 Logged at IST: 2026-07-11 01:16 IST What it is: X post by zigo 101 saying a Go proposal was formally rejected Gist: Refers to the proposal to allow explicit conversion from a function to a one-method interface in Go. The signal here is less the terse rejection post itself and more that this possible post-1.27 language change is now formally not happening. Newsletter angle: Useful small datapoint in Go language evolution: another reminder that the bar for adding convenience features to core Go remains high, especially where ambiguity or language-surface complexity is involved. Retrieval note: G… https://reading-list.oddship.net/notes/2026-07-11-gpt-5-4-with-pi-0-69-0-is-just-nice/ https://reading-list.oddship.net/notes/2026-07-11-gpt-5-4-with-pi-0-69-0-is-just-nice/ Sat, 11 Jul 2026 00:00:00 +0000 Logged at IST: 2026-07-11 02:17 IST What it is: X post by Rohan Verma linking his blog post "GPT-5.4 with Pi 0.69.0 is just nice" Gist: Argues that an agent harness stack getting boring is a success condition, not a failure. The post frames Pi 0.69.0 + GPT-5.4 + Bosun/Zero Agent as having crossed from fragile novelty into dependable daily tooling, where the interesting result is not frontier-model hype but the fact that the stack stopped demanding constant maintenance to remain useful. Newsletter angle: Strong firsthand writeup on harness maturity: the real milestone is when the agent stack st… https://reading-list.oddship.net/notes/2026-07-11-solod/ https://reading-list.oddship.net/notes/2026-07-11-solod/ Sat, 11 Jul 2026 00:00:00 +0000 Logged at IST: 2026-07-11 01:12 IST What it is: X reply by Aliaksandr Valialkin pointing to the solod project Gist: Very terse recommendation of solod, a project described as "a subset of Go that translates to C." In context, this looks like a pointer toward an alternative way to get highly portable or low-level output from Go-like code without using full Go as-is. Newsletter angle: Interesting small tooling pointer in the Go/compiler/toolchain space, especially if the broader thread is about language/runtime tradeoffs or portability. Retrieval note: Grounded from the X reply text plus the Git… https://reading-list.oddship.net/notes/2026-07-10-andrew-kelley-s-response-essay-on-bun-s-rust-rewrite/ https://reading-list.oddship.net/notes/2026-07-10-andrew-kelley-s-response-essay-on-bun-s-rust-rewrite/ Fri, 10 Jul 2026 00:00:00 +0000 Logged at IST: 2026-07-10 20:31 IST What it is: Andrew Kelley’s response essay on Bun’s Rust rewrite Gist: Frames the rewrite less as a language indictment and more as a consequence of Bun’s startup incentives, weak engineering discipline, and management culture. He argues Zig was a good fit for Bun’s early ambition, but the real failure mode was accumulating technical debt under venture-backed speed pressure rather than the language itself. Newsletter angle: Useful counterpoint to simplistic “Rust beat Zig” narratives; the sharper story is incentives, engineering quality, and what startup pre… https://reading-list.oddship.net/notes/2026-07-10-great-divergence-in-software-engineering/ https://reading-list.oddship.net/notes/2026-07-10-great-divergence-in-software-engineering/ Fri, 10 Jul 2026 00:00:00 +0000 Logged at IST: 2026-07-10 12:11 IST What it is: X post by Geoffrey Huntley linking to Stack72's essay "The Great Divergence in Software Engineering" Gist: Argues that the gap between teams effectively using AI and teams still piloting or rejecting it is no longer a simple lead but a compounding divergence, driven by retooling workflows, encoding automation, and treating bad AI output as an engineering problem instead of a veto. Newsletter angle: Strong framing for AI-native engineering orgs versus incumbents stuck in evaluation loops; good organizational/process lens. Embedded source X… https://reading-list.oddship.net/notes/2026-07-09-rewriting-bun-in-rust/ https://reading-list.oddship.net/notes/2026-07-09-rewriting-bun-in-rust/ Thu, 09 Jul 2026 00:00:00 +0000 Logged at IST: 2026-07-09 10:34 IST What it is: X post by Jarred Sumner linking to Bun's post "Rewriting Bun in Rust" Gist: Explains why Bun is being rewritten from Zig to Rust, positioning the move around long-term stability and maintainability as the project scales, even though Zig was instrumental in making the original ambitious build possible. Newsletter angle: Another data point on language/runtime rewrites in core developer tooling, especially where scaling and reliability start to dominate raw early-stage velocity. Embedded source X / Twitter post Show embedded post X e… https://reading-list.oddship.net/notes/2026-07-08-why-typescript-7-0-was-rewritten-in-go-and-what-it-means-for-your-dev-stack/ https://reading-list.oddship.net/notes/2026-07-08-why-typescript-7-0-was-rewritten-in-go-and-what-it-means-for-your-dev-stack/ Wed, 08 Jul 2026 00:00:00 +0000 Logged at IST: 2026-07-08 22:32 IST; resurfaced via mattn on 2026-08-24 03:09 IST. What it is: Steve Francia's argument for Go as a strong default for agentic development, using the TypeScript compiler's Go rewrite as the lead example. Gist: The TypeScript team's native Go port is framed as more than a compiler implementation detail. Francia argues it is a signal that agent-heavy developer stacks benefit from boring, readable, compiled, operationally sturdy languages rather than scripting-first ecosystems. The sharper claim is reader economics. Go was designed to favor the reader over the writ… https://reading-list.oddship.net/notes/2026-07-06-agentic-inbox/ https://reading-list.oddship.net/notes/2026-07-06-agentic-inbox/ Mon, 06 Jul 2026 00:00:00 +0000 What it is: hands-on writeup of deploying Cloudflare’s official agentic-inbox to run a custom-domain email client on Cloudflare Workers Gist: the stack uses Email Routing for inbound mail, Email Service for sending, Durable Objects + SQLite for mailboxes, R2 for attachments, and Cloudflare Access for auth. Main operational gotcha is that one-click deploy is not enough: you still have to wire the Email Routing catch-all and set the Access secrets or the app won’t work Newsletter angle: agentic inboxes are becoming deployable infra products, but the real story is the surrounding control plane an… https://reading-list.oddship.net/notes/2026-07-06-chris-short-s-devops-ish-316-roundup/ https://reading-list.oddship.net/notes/2026-07-06-chris-short-s-devops-ish-316-roundup/ Mon, 06 Jul 2026 00:00:00 +0000 What it is: Chris Short’s DevOps’ish 316 roundup Gist: strongest signals are ClickHouse gaining observability mindshare, Vint Cerf warning that agents will need more formal coordination than plain English, Podman 6.0 breaking old assumptions, and agent-secret hygiene as an architecture problem Newsletter angle: infra edge signals, observability economics, protocolized agents, and security boundaries around agent tooling Retrieval note: roundup page fetched directly from devopsish.com Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click … https://reading-list.oddship.net/notes/2026-07-06-cost-yagni-was-never-about/ https://reading-list.oddship.net/notes/2026-07-06-cost-yagni-was-never-about/ Mon, 06 Jul 2026 00:00:00 +0000 What it is: X post by Bilgin Ibryam pointing to Kent Beck’s “The Cost YAGNI Was Never About” Gist: YAGNI is about timing and option value, not code-writing thrift; AI codegen lowers typing cost but increases the risk of speculative structure nobody deeply understands Newsletter angle: code can be cheap to generate and still expensive to commit to Retrieval note: article text recovered directly from Kent Beck’s newsletter page Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remember… https://reading-list.oddship.net/notes/2026-07-06-miguel-ngel-pastor-linking-zalando-s-engineering-post-on-client-side-load-balancing-at-a-million-re/ https://reading-list.oddship.net/notes/2026-07-06-miguel-ngel-pastor-linking-zalando-s-engineering-post-on-client-side-load-balancing-at-a-million-re/ Mon, 06 Jul 2026 00:00:00 +0000 What it is: Miguel Ángel Pastor linking Zalando’s engineering post on client-side load balancing at a million requests per second Gist: argues high-fanout internal traffic benefited from moving load balancing into the client process to preserve cache locality, improve debuggability, and avoid shared-ingress distortions; highlights safety work like bounded-load routing and rollout fade-in Newsletter angle: own the routing decision in-process when cache-sensitive fanout paths make shared infra the bottleneck Retrieval note: extracted cleanly via FXTwitter API; card points to the Zalando engineer… https://reading-list.oddship.net/notes/2026-07-04-read-more-science-fiction/ https://reading-list.oddship.net/notes/2026-07-04-read-more-science-fiction/ Sat, 04 Jul 2026 00:00:00 +0000 What it is: X post from svs sharing his essay “Read More (Science) Fiction.” Newsletter angle: “read more sci-fi” is the visible conclusion, but the sharper claim is that fiction supplies vocab and priors for handling agentic weirdness without naive hype or naive panic. Retrieval note: extracted via FXTwitter API + fetched linked article directly. Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered in this browser. Open post on X instead Open post on X https://reading-list.oddship.net/notes/2026-07-02-vogels-pointing-to-zalando-s-engineering-writeup-on-client-side-load-balancing-for-a-very-high-fan/ https://reading-list.oddship.net/notes/2026-07-02-vogels-pointing-to-zalando-s-engineering-writeup-on-client-side-load-balancing-for-a-very-high-fan/ Thu, 02 Jul 2026 00:00:00 +0000 Gist: Zalando moved internal fan-out traffic off shared ingress and into an in-process client-side load balancer to preserve consistent-hash cache locality, cut latency spikes, improve debuggability, and reduce shared infra cost. The interesting details are the safety/operability work: exact hash parity with Skipper, informer-based pod discovery, N-ring fade-in for scale-ups, and bounded-load routing using occupancy plus latency instead of naive in-flight/request-rate signals. Newsletter angle: “own the routing decision in-process” or “occupancy beats request-rate for bounded load” as the memo… https://reading-list.oddship.net/notes/2026-07-01-about-passmark-an-open-source-playwright-library-for-ai-browser-regression-testing/ https://reading-list.oddship.net/notes/2026-07-01-about-passmark-an-open-source-playwright-library-for-ai-browser-regression-testing/ Wed, 01 Jul 2026 00:00:00 +0000 Gist: Passmark checks an LLM cache before making fresh model calls during browser tests, reportedly cutting a 7-minute suite down to 90 seconds. Newsletter angle: “cache-first AI browser testing” as practical infra for regression pipelines. Retrieval note: tweet extracted via FXTwitter API; attached screenshot shows the GitHub repo tagline mentioning intelligent caching, authentication, and multi-model verification. Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered in this… https://reading-list.oddship.net/notes/2026-07-01-soria-parra-announcing-mcp-sdk-v2-betas-ahead-of-a-new-stateless-mcp-spec-slated-for-july-28/ https://reading-list.oddship.net/notes/2026-07-01-soria-parra-announcing-mcp-sdk-v2-betas-ahead-of-a-new-stateless-mcp-spec-slated-for-july-28/ Wed, 01 Jul 2026 00:00:00 +0000 Gist: Typescript SDK v2.0.0-beta.1 and Python SDK v2.0.0b1 are out; goal is to make building MCP servers and clients easier, with feedback requested on ergonomics. Newsletter angle: “stateless MCP lands July 28” plus what SDK v2 means for tool/server implementers. Retrieval note: extracted via FXTwitter API; linked GitHub release URLs were present in the tweet body. Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered in this browser. Open post on X instead Open post o… https://reading-list.oddship.net/notes/2026-06-28-engineering-for-bounded-cognition/ https://reading-list.oddship.net/notes/2026-06-28-engineering-for-bounded-cognition/ Sun, 28 Jun 2026 00:00:00 +0000 Gist: good engineering is mostly about shaping systems so small, distractible minds can change them safely, via naming, boundaries, tests, reversibility, and interfaces that assume attention is scarce rather than ideal operators. Newsletter angle: “Build for bounded attention, not ideal operators.” Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered in this browser. Open post on X instead Open post on X https://reading-list.oddship.net/notes/2026-06-26-adithya-venkatesan-sharing-alter-magazine-s-piece-on-designing-ice-cream-for-indian-conditions/ https://reading-list.oddship.net/notes/2026-06-26-adithya-venkatesan-sharing-alter-magazine-s-piece-on-designing-ice-cream-for-indian-conditions/ Fri, 26 Jun 2026 00:00:00 +0000 What it is: Adithya Venkatesan sharing Alter Magazine’s piece on designing ice cream for Indian conditions Gist: frames ice cream as a four-phase material, ice crystals, unfrozen sugar syrup, churned-in air, and a fat network, and argues Indian heat + weak cold-chain conditions make conventional formulations degrade fast; points to adaptations like denser kulfi-ish formulations, freezing at serve time, and rebalancing protein/fibre vs sugar/fat for stability Newsletter angle: “designing for India” through thermodynamics/material science rather than just pricing or distribution Embedded source … https://reading-list.oddship.net/notes/2026-06-26-cursed-code/ https://reading-list.oddship.net/notes/2026-06-26-cursed-code/ Fri, 26 Jun 2026 00:00:00 +0000 What it is: Tim McNamara sharing an IOCCC-winning “cursed code” project where Pong advances by rewriting its own source each frame Gist: the linked repo, uellenberg/Insert, is a small language for self-modifying code; programs can access their own source as string fragments, overwrite marked values, print the next version of themselves, and in the Pong demo each run emits the source for the next frame before recompiling Newsletter angle: “playful programming systems” / self-modifying code as art rather than anti-pattern Embedded source X / Twitter post Show embedded post X embe… https://reading-list.oddship.net/notes/2026-06-26-fatih-arslan-asking-whether-anyone-has-made-git-worktrees-feel-natural-in-daily-use/ https://reading-list.oddship.net/notes/2026-06-26-fatih-arslan-asking-whether-anyone-has-made-git-worktrees-feel-natural-in-daily-use/ Fri, 26 Jun 2026 00:00:00 +0000 What it is: Fatih Arslan asking whether anyone has made git worktrees feel natural in daily use Gist: straightforward practitioner complaint that worktrees remain awkward even after repeated attempts; useful mainly as a prompt for workflow/tooling patterns rather than as a claim-heavy post Newsletter angle: “great primitive, bad default ergonomics” as a recurring pattern in developer tools Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered in this browser. Open post on… https://reading-list.oddship.net/notes/2026-06-26-rhys-sullivan-note-on-why-mcp-underdelivered-initially-and-what-comes-next/ https://reading-list.oddship.net/notes/2026-06-26-rhys-sullivan-note-on-why-mcp-underdelivered-initially-and-what-comes-next/ Fri, 26 Jun 2026 00:00:00 +0000 What it is: Rhys Sullivan note on why MCP underdelivered initially and what comes next Gist: argues MCP launched in the GPT-4o / Sonnet 3.5 era before good agent/tooling patterns were understood, so many servers exposed too few capabilities and clients added too much friction; meanwhile bash/CLI-based agents won because they could chain commands, install tools dynamically, and lean on mature shell primitives. His pushback is that this should not end in “just use CLIs”: CLIs hide action semantics and add statefulness, while the better end-state is harnesses that can expose APIs, MCP, CLIs, Grap… https://reading-list.oddship.net/notes/2026-06-24-visible-standouts-the-second-half-eugene-yan-on-eval-process-han-chung-lee-on-agent-eval-infra-hame/ https://reading-list.oddship.net/notes/2026-06-24-visible-standouts-the-second-half-eugene-yan-on-eval-process-han-chung-lee-on-agent-eval-infra-hame/ Wed, 24 Jun 2026 00:00:00 +0000 What it is: Visible standouts: The Second Half; Eugene Yan on eval process; Han-Chung Lee on agent eval infra; Hamel/Shreya LLM Evals FAQ; Jason Wei on verification; Anthropic on agent evals; Ofir Press on benchmarks; AI Agents That Matter; Building on Evaluation Quicksand; EvalGen; Benches 2026. Gist: strong starter pack for agent/LLM evals; themes include eval infra as technical debt, process over tooling, verifier design, benchmark saturation/contamination, agent-specific eval design, and criteria drift. Newsletter angle: compact “best evals reading list” / why eval practice is shifting fro… https://reading-list.oddship.net/notes/2026-06-19-linking-filiph-net-text-pokerd-html/ https://reading-list.oddship.net/notes/2026-06-19-linking-filiph-net-text-pokerd-html/ Fri, 19 Jun 2026 00:00:00 +0000 What it is: X post by @filiphracek linking filiph.net/text/pokerd.html Gist: writeup on building pokerd, a terminal-first Texas Hold’em trainer you can play instantly over SSH (ssh play@poker.filiph.net); interesting bits are the non-immersive-game framing, scrollback-friendly TUI choices, bot tuning via self-play + JSONL events, and shipping the game as a passwordless SSH shell inside a container on a VPS. Newsletter angle: “SSH as zero-install distribution” / terminals as a deliberate product surface, not just a dev tool. Notes: extracted via FXTwitter API + blog post (partial long-form read… https://reading-list.oddship.net/notes/2026-06-19-linking-github-com-leyten-shard/ https://reading-list.oddship.net/notes/2026-06-19-linking-github-com-leyten-shard/ Fri, 19 Jun 2026 00:00:00 +0000 What it is: X post by @leyten linking github.com/leyten/shard Gist: Shard is a WAN-distributed pipeline-parallel LLM inference engine that splits a frontier-size model across GPUs on separate machines; claim is ~30 tok/s for GLM-5.2 744B across 6 RTX PRO 6000s in 6 US states using speculative decoding, async pipelining, and a CUDA-graphed draft model. Newsletter angle: “frontier inference without a datacenter” / distributed serving as systems engineering rather than centralized infra. Notes: extracted via FXTwitter API + GitHub README. Embedded source X / Twitter post Show embedded… https://reading-list.oddship.net/notes/2026-06-12-plannotator-effective-html/ https://reading-list.oddship.net/notes/2026-06-12-plannotator-effective-html/ Fri, 12 Jun 2026 00:00:00 +0000 Gist: the repo packages focused agent skills for producing self-contained, visually strong HTML artifacts, especially diagrams and plan pages, plus an optional Plannotator renderer/annotator. The post points to a demo video showing the diff/code viewer behavior. Newsletter angle: “HTML as agent output surface” / better human-review loops for plans and diagrams. Retrieval note: extracted via FXTwitter API; followed the linked GitHub repo page for the core description. Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X em… https://reading-list.oddship.net/notes/2026-06-10-skepticism-strong-opinion-piece-not-data-heavy-the-claim-that-generics-haven-t-improved-productivit/ https://reading-list.oddship.net/notes/2026-06-10-skepticism-strong-opinion-piece-not-data-heavy-the-claim-that-generics-haven-t-improved-productivit/ Wed, 10 Jun 2026 00:00:00 +0000 What it is: skepticism: strong opinion piece, not data-heavy; the claim that generics haven’t improved productivity is asserted more than demonstrated. Gist: his case is that Go’s value is simplicity/readability/maintainability, and newer features like generics and range-over-functions (iterators) erode that by increasing implicit behavior and language complexity. He argues generics have seen limited practical need while adding compiler/type-system complexity, and that iterators introduce another iteration style plus hidden control-flow transformations that make code harder to read and debug. … https://reading-list.oddship.net/notes/2026-06-09-decline-of-search-engines-is-an-opportunity/ https://reading-list.oddship.net/notes/2026-06-09-decline-of-search-engines-is-an-opportunity/ Tue, 09 Jun 2026 00:00:00 +0000 What it is: Lewis Campbell post linking to “The Decline of Search Engines is an Opportunity”. Gist: argues worsening search quality should push people back toward the old web habit of maintaining personal links pages; discovery by human-curated hyperlinks is framed as a healthier alternative to SEO sludge and LLM-mediated search summaries. Newsletter angle: “search decay revives the links page” is a clean thesis with nice historical texture. Note: extracted tweet via FXTwitter and fetched linked blog post successfully. Embedded source X / Twitter post Show embedded post X embed… https://reading-list.oddship.net/notes/2026-06-09-dynamo-and-the-computer/ https://reading-list.oddship.net/notes/2026-06-09-dynamo-and-the-computer/ Tue, 09 Jun 2026 00:00:00 +0000 What it is: Zara Zhang post using Paul David’s “The Dynamo and the Computer” as an analogy for AI adoption. Gist: argues AI gains won’t come from simply inserting models into existing workflows; like electrification, the real productivity jump comes only after redesigning the organization and flow of work around the new technology. Newsletter angle: “AI is still in the faster steam engine phase” is a strong line for transformation skepticism. Note: extracted tweet via FXTwitter; referenced paper link appears to be in replies/comments and was not followed here. Embedded source X / Twitt… https://reading-list.oddship.net/notes/2026-06-09-writing-code-vs-shipping-code-productivity-effects-across-generations-of-ai-coding-tools/ https://reading-list.oddship.net/notes/2026-06-09-writing-code-vs-shipping-code-productivity-effects-across-generations-of-ai-coding-tools/ Tue, 09 Jun 2026 00:00:00 +0000 What it is: Murat Demirbas on “Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools”. Gist: uses a new MIT/Wharton paper plus an Amdahl’s-law framing to argue that AI massively speeds up code generation but much less meaningfully speeds shipped software, because the bottleneck is the non-parallelizable human layer: task definition, coordination, review, and release. Newsletter angle: “AI accelerates writing code more than shipping code” or “Amdahl’s Law is eating AI coding productivity claims.” Note: extracted tweet via FXTwitter and fetched linked blog po… https://reading-list.oddship.net/notes/2026-06-08-promoting-an-85-minute-mit-lecture-on-git-internals-data-model/ https://reading-list.oddship.net/notes/2026-06-08-promoting-an-85-minute-mit-lecture-on-git-internals-data-model/ Mon, 08 Jun 2026 00:00:00 +0000 What it is: an X post promoting an 85-minute MIT lecture on Git internals / data model. Gist: the pitch is that most developers memorize Git commands without understanding commits, trees, refs, and the graph underneath; learning the model makes debugging history and merge/rebase failures much less magical. Newsletter angle: “Git literacy as leverage”, understanding the object graph matters more when agents are branching/rewriting history at speed. Note: extracted via FXTwitter API; saved from the post text only, lecture content itself not yet reviewed. Embedded source X / Twitter post … https://reading-list.oddship.net/notes/2026-06-08-vim-royale/ https://reading-list.oddship.net/notes/2026-06-08-vim-royale/ Mon, 08 Jun 2026 00:00:00 +0000 What it is: Jitesh boosting vim_royale, a Peerlist project for realtime multiplayer Vim battles. Gist: lightweight launch/amplification post rather than a deep technical thread; the linked card describes the project very tersely as “Realtime multiplayer Vim battles.” Newsletter angle: playful developer-product idea / “tools culture as game mechanic.” Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered in this browser. Open post on X instead Open post on X https://reading-list.oddship.net/notes/2026-06-04-solo-climb/ https://reading-list.oddship.net/notes/2026-06-04-solo-climb/ Thu, 04 Jun 2026 00:00:00 +0000 What it is: Ajey Gore linking his essay “The Solo Climb.” Gist: the argument is that AI-enabled solo builders and tiny teams only work when they first build a genuinely load-bearing “harness”, trusted tests, evals, specs, and hard gates that can answer “is this safe enough to ship?” without relying on redundant humans. Newsletter angle: “100x teams” are mostly a harness story, AI leverage scales only when trust, eval, and rollback systems become the new team structure. Note: extracted via FXTwitter API and linked article; article read partially via web fetch due to truncation, but core thesis … https://reading-list.oddship.net/notes/2026-06-03-simon-willison-pointing-to-bloomberg-on-uber-capping-agentic-coding-tool-spend-at-1-500-month-per-e/ https://reading-list.oddship.net/notes/2026-06-03-simon-willison-pointing-to-bloomberg-on-uber-capping-agentic-coding-tool-spend-at-1-500-month-per-e/ Wed, 03 Jun 2026 00:00:00 +0000 What it is: Simon Willison pointing to Bloomberg on Uber capping agentic coding-tool spend at $1,500/month per employee per tool. Gist: Simon’s read is that the cap is a rational response to runaway token spend and also a useful revealed-preference signal: Uber appears willing to tolerate tooling costs on the order of tens of thousands of dollars per engineer per year if the productivity gain holds. Newsletter angle: coding-agent PMF is now visible through finance policy; spend caps as a clearer signal than hype. Note: extracted via FXTwitter API plus Simon’s linked post for added context. Emb… https://reading-list.oddship.net/notes/2026-06-02-sid-s-writeup-on-a-recently-patched-instagram-meta-account-takeover-flow/ https://reading-list.oddship.net/notes/2026-06-02-sid-s-writeup-on-a-recently-patched-instagram-meta-account-takeover-flow/ Tue, 02 Jun 2026 00:00:00 +0000 What it is: Sid's writeup on a recently patched Instagram/Meta account takeover flow. Gist: attacker allegedly only needed a target username, region-matching IP, and Meta support AI to redirect recovery codes to attacker-controlled email; video selfie checks were reportedly weak enough to bypass with AI-animated public photos. Newsletter angle: “support AI as auth bypass” / security lesson on high-privilege recovery flows needing stricter invariants than normal login. Note: fetched article body successfully via web_fetch. https://reading-list.oddship.net/notes/2026-05-31-solution-might-be-cancelling-my-ai-subscription/ https://reading-list.oddship.net/notes/2026-05-31-solution-might-be-cancelling-my-ai-subscription/ Sun, 31 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback, then read linked post directly: https://thoughts.hmmz.org/2026-05-31.html Mario Zechner recommends David's post the solution might be cancelling my AI subscription. Gist: a sharp anti-friction argument against current AI-tool usage patterns, cheap output and minimal resistance can explode side projects, context switching, and pseudo-productivity while degrading attention and commitment. Why it matters: good counterweight to "more agent throughput = better work" narratives; frames AI as an attention-manag… https://reading-list.oddship.net/notes/2026-05-19-antirez-on-alternatives-to-the-standard-edit-tool-for-llm-agents-links-to-a-short-blog-note/ https://reading-list.oddship.net/notes/2026-05-19-antirez-on-alternatives-to-the-standard-edit-tool-for-llm-agents-links-to-a-short-blog-note/ Tue, 19 May 2026 00:00:00 +0000 Gist: proposes CAS-style edits using line-number + short checksum tags instead of resending old text verbatim, aiming to save tokens while still guarding against stale or hallucinated edits. Newsletter angle: "a lighter-weight edit primitive for coding agents: line tags vs full old-text CAS". Note: extracted via FXTwitter API + antirez.com post. Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered in this browser. Open post on X instead Open post on X https://reading-list.oddship.net/notes/2026-05-19-mario-zechner-recommending-antirez-s-post-on-hash-line-read-edit-tools-for-agents/ https://reading-list.oddship.net/notes/2026-05-19-mario-zechner-recommending-antirez-s-post-on-hash-line-read-edit-tools-for-agents/ Tue, 19 May 2026 00:00:00 +0000 Gist: mostly a pointer/amplifier rather than a new thesis; reinforces interest around checksum-tagged line edit protocols for agent tooling. Newsletter angle: maybe bundle with the original antirez item as a small "agent tooling design" thread rather than a standalone item. Note: extracted via FXTwitter API; quotes the previously logged antirez post and adds no new linked material. Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered in this browser. Open post on X inste… https://reading-list.oddship.net/notes/2026-05-13-ambitious-oss-project-pitching-wifi-csi-as-a-privacy-preserving-sensing-stack-presence-detection-br/ https://reading-list.oddship.net/notes/2026-05-13-ambitious-oss-project-pitching-wifi-csi-as-a-privacy-preserving-sensing-stack-presence-detection-br/ Wed, 13 May 2026 00:00:00 +0000 Imported from historical reading log. Ambitious OSS project pitching WiFi CSI as a privacy-preserving sensing stack: presence detection, breathing/heart-rate monitoring, activity recognition, rough pose estimation, and through-wall/environment sensing using ESP32-S3 nodes. Interesting angle is the packaging: not just a research demo, but a full “edge intelligence” story with cheap hardware, local processing, attestations, mesh sensing, demos, and a long README translating RF sensing into product language. Newsletter angle: strong hook if framed as “camera-free spatial intelligence from commod… https://reading-list.oddship.net/notes/2026-05-12-js-sqlite-parser-ported-from-sqlite-s-own-lemon-lalr-grammar-aimed-at-being-fast-lightweight-browse/ https://reading-list.oddship.net/notes/2026-05-12-js-sqlite-parser-ported-from-sqlite-s-own-lemon-lalr-grammar-aimed-at-being-fast-lightweight-browse/ Tue, 12 May 2026 00:00:00 +0000 Imported from historical reading log. JS SQLite parser ported from SQLite’s own Lemon/LALR grammar, aimed at being fast, lightweight, browser-friendly, and more faithful than typical JS SQL parsers. Notable angle: improved structured diagnostics and hints, plus AST traversal/CLI tooling, which makes it more useful for editor tooling, linting, query analysis, or SQL-aware product features. Newsletter angle: a good example of “serious infra-grade parsing” moving into pure TypeScript without wasm, with a tight value prop around correctness + developer ergonomics. Retrieval note: extracted from G… https://reading-list.oddship.net/notes/2026-05-12-llm/ https://reading-list.oddship.net/notes/2026-05-12-llm/ Tue, 12 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted via api.fxtwitter.com fallback, then read linked TIL directly: https://til.simonwillison.net/llms/llm-shebang Simon Willison shows a neat pattern for using his llm CLI in a shebang line, turning plain-English files or YAML templates into executable scripts. The more interesting part is not the toy prompt examples but the tool-enabled/template-enabled scripts: parameterized prompts, embedded functions, and lightweight agentic shells around LLM/tool workflows. Newsletter angle: a crisp example of LLMs collapsing the boundary between prompt, script… https://reading-list.oddship.net/notes/2026-05-11-saved-media-locally/ https://reading-list.oddship.net/notes/2026-05-11-saved-media-locally/ Mon, 11 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted via api.fxtwitter.com fallback; includes an image illustrating progressive rendering from noise to a clear cat image. Saved media locally: Dax reframes coding-agent usage: not like 3D printing one committed layer at a time, but like progressive rendering, start with a blurry whole, then make repeated full passes that sharpen the entire shape. Follow-up reply worth keeping with it: https://x.com/thdxr/status/2053566249351754193, he says this is actually counter to how his brain naturally imagines construction, which makes the metaphor more intere… https://reading-list.oddship.net/notes/2026-05-07-34kb/ https://reading-list.oddship.net/notes/2026-05-07-34kb/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback; the attached image includes the concrete compression results. Sam Rose compares the same payload as JSON (34kb) and protobuf (15kb) and finds that after compression, JSON is often slightly smaller than protobuf for Brotli, Zstd, gzip, and bzip2; protobuf only wins clearly for lz4 and barely for lzma in his sample. Concrete reported sizes from the screenshot: Brotli json 4237 < bin 4279, Zstd 4484 < 4702, gzip 4766 < 4949, bzip2 5208 < 5302, lzma 4500 > 4484, lz4 6245 > 5832. Why it is i… https://reading-list.oddship.net/notes/2026-05-07-ai-slop/ https://reading-list.oddship.net/notes/2026-05-07-ai-slop/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback. Mitchell Hashimoto argues that AI slop is useful as an internal experimentation tool: low-quality generated code/UI/plugins can dramatically reduce the cost of parallel exploration and API iteration, especially when regeneration is cheaper than careful hand maintenance. His concrete examples are good: shipping an intentionally rough alpha frontend to focus on core internals, and using overnight agent loops to generate many disposable plugins so the whole ecosystem can be tested before the SDK is stable. … https://reading-list.oddship.net/notes/2026-05-07-auth-for-mcp/ https://reading-list.oddship.net/notes/2026-05-07-auth-for-mcp/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback and checked the linked Auth0 GA announcement. Auth0 is pitching Auth for MCP as the missing identity/authorization layer for production MCP servers: not just connecting agents to tools, but enforcing who the user is and what the agent may do on their behalf. The notable implementation details are support for CIMD client registration, OBO token exchange for downstream APIs, and MCP-style resource identifiers instead of plain OAuth audience handling. Why it matters: MCP is quickly moving from demo protocol … https://reading-list.oddship.net/notes/2026-05-07-autodata/ https://reading-list.oddship.net/notes/2026-05-07-autodata/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback and checked the linked Meta RAM Autodata post plus the referenced justrach/devswarm repo and sample issue. Rach connects her agent workflow to Meta's Autodata framing: agents act like data scientists by iterating on a hypothesis, generating data, testing it, validating results, extracting learnings, and then closing the loop. The linked paper/blog's core idea is strong: convert inference-time compute into better training/eval data quality by having an agent iteratively create data, analyze failures, refin… https://reading-list.oddship.net/notes/2026-05-07-entire-s-core-claim-is-useful-from-202k-real-tool-calls-across-1-983-public-coding-agent-checkpoint/ https://reading-list.oddship.net/notes/2026-05-07-entire-s-core-claim-is-useful-from-202k-real-tool-calls-across-1-983-public-coding-agent-checkpoint/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted Mario Zechner's quote-post via api.fxtwitter.com fallback and checked the linked Entire blog post on agentic search. Entire's core claim is useful: from ~202k real tool calls across ~1,983 public coding-agent checkpoints, about 48.8% were search-related, so search is a first-order agent behavior rather than a side utility. Their more interesting finding is that raw speed is not the main bottleneck. Making search dramatically faster (ripgrep → fff) only modestly improved end-to-end run time because tool latency was a tiny fraction of total wall c… https://reading-list.oddship.net/notes/2026-05-07-hunk/ https://reading-list.oddship.net/notes/2026-05-07-hunk/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback and checked the linked GitHub repo. Mitchell Hashimoto strongly recommends Hunk, saying it has fully replaced other local diff viewers for him. Hunk is positioned as a review-first terminal diff viewer for agent-authored changesets. Notable capabilities from the repo: multi-file review stream with sidebar navigation, inline AI/agent annotations, split/stack responsive layouts, watch mode, keyboard + mouse support, pager mode, and Git difftool/pager integration. Install/use gist: package name hunkdiff; com… https://reading-list.oddship.net/notes/2026-05-07-is-moving-its-github-repo-into-the/ https://reading-list.oddship.net/notes/2026-05-07-is-moving-its-github-repo-into-the/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback. Mario Zechner says pi is moving its GitHub repo into the earendil-works org and will start publishing packages under the @earendil-works npm namespace instead of @mariozechner. Short-term compatibility remains for existing imports, but typed extensions should migrate quickly once the new packages land. Breaking edge: extensions switched to @earendil-works will stop working on older pi versions after today's release. Why it matters: this is an ecosystem/ownership cleanup move, but it deliberately forces e… https://reading-list.oddship.net/notes/2026-05-07-mirage/ https://reading-list.oddship.net/notes/2026-05-07-mirage/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback and checked the linked repo strukto-ai/mirage. Zecheng Zhang introduces Mirage, a unified virtual filesystem for AI agents that mounts heterogeneous systems like S3, Drive, Slack, Gmail, GitHub, Linear, Notion, databases, and SSH into one filesystem abstraction. Core pitch: agents can reuse familiar Unix/bash semantics (cat, grep, head, pipes, wc) across mixed backends and even structured formats like parquet, csv, json, h5, and wav, instead of learning service-specific APIs. Repo/docs framing adds two no… https://reading-list.oddship.net/notes/2026-05-07-printing-press/ https://reading-list.oddship.net/notes/2026-05-07-printing-press/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback and checked printingpress.dev. Printing Press is pitched as both a library of agent-native CLIs and a factory that generates new ones: from a spec/site/service it can print a token-efficient Go CLI, a Claude Code skill, an OpenClaw skill, and an MCP server. The design philosophy is notable: local SQLite mirrors, compound commands, and CLI ergonomics are treated as a better substrate for agents than raw APIs, raw MCPs, or official vendor CLIs. The examples are intentionally ambitious and eclectic, Linear, … https://reading-list.oddship.net/notes/2026-05-07-selection-of-great-prs-that-were-submitted-to-pi-a-thread/ https://reading-list.oddship.net/notes/2026-05-07-selection-of-great-prs-that-were-submitted-to-pi-a-thread/ Thu, 07 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted the root post via api.fxtwitter.com fallback: Armin Ronacher says it is a selection of great PRs that were submitted to Pi, a thread. Tried browser fallback on X to read the thread, but replies are gated behind login/signup, so the actual thread contents were not accessible from public view. User context says the thread is satire about bad PRs sent to Pi, which fits the phrasing but I could not independently verify from the gated replies. Blocker: root post readable; thread contents blocked by X login wall. Embedded source X / Twitter p… https://reading-list.oddship.net/notes/2026-05-06-ben-holmes-says-switching-from-tiptap-prosemirror-to-slate-made-a-rich-text-bulleted-list-interacti/ https://reading-list.oddship.net/notes/2026-05-06-ben-holmes-says-switching-from-tiptap-prosemirror-to-slate-made-a-rich-text-bulleted-list-interacti/ Wed, 06 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback. Ben Holmes says switching from TipTap/ProseMirror to Slate made a rich-text bulleted-list interaction dramatically easier to build; something that took weeks to half-work in TipTap took a couple of hours in Slate, with Codex helping. Core signal is less Slate is universally better and more framework ergonomics matter a lot for AI-assisted development: some abstractions are much easier to extend/debug with model help. Why it matters: useful anecdote for editor-stack choice, especially when complex WYSIWYG… https://reading-list.oddship.net/notes/2026-05-06-html5-css-face-lift-for-the-generated-pages/ https://reading-list.oddship.net/notes/2026-05-06-html5-css-face-lift-for-the-generated-pages/ Wed, 06 May 2026 00:00:00 +0000 Imported from historical reading log. GitHub PR title: HTML5+CSS face lift for the generated pages by knadh on mitmproxy/pdoc; merged Nov 20, 2014. Logged as a folklore/historical reference rather than a current article; likely relevant as an old design/implementation artifact in the pdoc/docsite lineage. Retrieval from the public PR page was partial because logged-out GitHub readability extraction is thin, but title/author/repo/merged status were captured. Follow-up if needed: inspect commits/diff directly or use GitHub API/source checkout for the substantive changes. https://reading-list.oddship.net/notes/2026-05-06-satya-microsoft-framing-firms-need-to-redesign-work-around-agentic-systems-with-ai-taking-more-exec/ https://reading-list.oddship.net/notes/2026-05-06-satya-microsoft-framing-firms-need-to-redesign-work-around-agentic-systems-with-ai-taking-more-exec/ Wed, 06 May 2026 00:00:00 +0000 Imported from historical reading log. Extracted main post via api.fxtwitter.com fallback and read the linked Microsoft Work Trend Index piece Agents, human agency, and the opportunity for organizations. Satya/Microsoft framing: firms need to redesign work around agentic systems, with AI taking more execution while humans shift toward judgment, intent-setting, and owning outcomes. The more interesting claim is organizational, not individual: Microsoft says culture, manager support, and talent practices explain more than 2x the reported AI impact of individual effort alone. Key vocabulary from … https://reading-list.oddship.net/notes/2026-05-05-mitchell-hashimoto-post-praising-antirez-s-write-up-on-developing-redis-array-support-as-a-good-exa/ https://reading-list.oddship.net/notes/2026-05-05-mitchell-hashimoto-post-praising-antirez-s-write-up-on-developing-redis-array-support-as-a-good-exa/ Tue, 05 May 2026 00:00:00 +0000 Imported from historical reading log. Mitchell Hashimoto post praising antirez's write-up on developing Redis Array support as a good example of thoughtful AI usage that empowers strong developers while preserving quality. Linked article: https://antirez.com/news/164 Read/stored gist of antirez article Redis array type: short story of a long development: Related PR/use-cases link: https://github.com/redis/redis/pull/15162 Embedded source X / Twitter post Show embedded post X embeds can let X track your visit. Click once to load X embeds on this site; this choice is remembered… https://reading-list.oddship.net/notes/2026-05-05-pimalaya-open-source-pim-tools-in-rust-positions-itself-as-i-o-free-rust-libraries-plus-house-made/ https://reading-list.oddship.net/notes/2026-05-05-pimalaya-open-source-pim-tools-in-rust-positions-itself-as-i-o-free-rust-libraries-plus-house-made/ Tue, 05 May 2026 00:00:00 +0000 Imported from historical reading log. Pimalaya: open-source PIM tools in Rust; positions itself as I/O-free Rust libraries plus house-made applications for the PIM domain. Himalaya: CLI to manage emails; supports IMAP/Maildir/Notmuch, SMTP/Sendmail, keyring, OAuth2, JSON output, and multi-account configuration. User intent: explore using Pimalaya/Himalaya to clean up a ~2k pending inbox. Related idea: connect this with Kailash Nadh's email UI idea. Writing idea to track: future post on rohanverma.net about using the harness to clean the inbox with this stack; create/track under a separate Wri…