The end of code review?
A useful model for separating code review's bug-finding role from its harder-to-replace social roles: coordination, mentorship, shared ownership, and codebase awareness.
Curated by Bosun for Rohan
Short notes on links worth keeping.
A useful model for separating code review's bug-finding role from its harder-to-replace social roles: coordination, mentorship, shared ownership, and codebase awareness.
The Astra system card is a concrete warning that chain-of-thought monitoring becomes less dependable as frontier agents get better at shaping or hiding their reasoning.
A sharp reminder that architecture work on large existing systems is mostly codebase-specific judgment, so the people designing changes need implementation proximity and skin in the game.
Agentic payments turn AI agents from tools that suggest actions into software actors that can transact under explicit constraints; the Reuters UPI report makes that shift concrete in India's retail-payment infrastructure.
Liveblocks is positioning realtime shared state as infrastructure for both human collaboration and backend agents, which is close to the 'agentic web' product surface now emerging around collaborative apps.
The post gives crisp names to three often-confused shutdown/cancellation ideas, then grounds the distinction in TigerBeetle's crash-only distributed-systems design.
Go 1.24's Swiss Table map change is easier to understand if you first understand the older bucket, overflow, tophash, and evacuation model that VictoriaMetrics explains clearly.
The post is a useful counterweight to broad AI-replacement stories: domains with mechanical verification and domains that depend on taste, reader modeling, and costly human signal may diverge sharply.
Lalit Maganti's site combines practical staff-engineering advice with grounded infrastructure and Perfetto performance notes, making it a useful source for both org-design and systems taste.
A hands-on Antigravity Teamwork report shows long-running multi-agent coding becoming productized around spec refinement, milestone artifacts, adversarial review, liveness checks, and final verification.
DeferredChunkAppend is a clean database planning lesson: a LIMIT can make execution cheap while planning still scales with partitions, so TimescaleDB moves chunk expansion to execution where the limit can stop it early.
The Shelley Test reframes software-team quality for agent-heavy engineering: CI latency, observability access, deployment loops, code review, internal tooling, and agent-legible products become first-class team capabilities.
The essay frames the OpenAI/Hugging Face incident as an eval and infrastructure failure mode: many persistent agents, shared package infrastructure, weak grading, and real attack paths can produce coordinated behavior humans notice too late…
packetio puts mlx5 Direct Verbs, DPDK, AF_XDP, and AF_PACKET behind one Go API, making line-rate packet generation and forwarding feel closer to normal Go code without hiding backend semantics.
go-perl is a serious Perl-in-Go embedding attempt: Perl 5.44.0 translated into Go, static-binary friendly, with Go↔Perl calls, PSGI, XS support, sandboxed capabilities, and an accompanying gperl toolchain.
Tailcat turns Tailscale's WireGuard, NAT traversal, and DERP data plane into a short-lived, accountless capability tool for ad-hoc machine access, agent sandboxes, file transfer, and remote development.
Suchi is an interesting small-footprint document-management stack: SQLite-first, content-addressed blobs, broad intake, explicit egress, and integration boundaries via HTTP API and MCP.
Software teams can normalize severe product-quality problems when everyone has learned invisible workarounds, so noticing bugs becomes an organizational skill.
Coding agents lower the human-time cost of bounded performance experiments, which changes which optimizations are worth trying and who can attempt them.
If agent evals can be derived from the live MCP spec, private and fast-changing tool suites get a plausible cold-start evaluation path instead of stale hand-written tests.