Murat Demirbas on why writing may be safer from AI
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.
Links: Original source · Shared link · Related link 1 · Related link 2 · Related link 3 · Related link 4 · Related link 5
Logged at IST: 2026-09-01 09:59 IST
What it is: Murat Demirbas's blog post arguing that writing may be one of the safer knowledge-work skills in the LLM era.
Gist: Demirbas's core claim is that LLMs are strongest where the feedback loop is tight and mechanical. Math has crisp specifications and binary verification. Code has compilers, tests, model checkers, and other executable checks. Structured domains such as law and finance have explicit frameworks and empirical data. Writing sits at the opposite end: no stable spec, no objective stopping rule, and no single ground truth for whether a piece works.
That makes prose a wicked, reader-dependent problem. Good writing requires continuously modeling what a specific reader knows, where their cognitive load is, and how each sentence will land. Demirbas argues that LLM prose still feels trapped in a slop plateau: fluent but generic, cadence-heavy, and missing lived experience or skin in the game.
The economics point is the useful part to keep. If AI can generate endless generic text, authentic human voice becomes a costly signal: proof that someone spent scarce attention, taste, and judgment on the piece. In that framing, writers do not need to optimize around AI nearly as much as programmers do; they were already working at the frontier where verification is human resonance.
Newsletter angle: A good counter-position to “AI eats all knowledge work”: the defensible human edge may be taste, reader modeling, and costly proof-of-work rather than raw generation.