antirez vs mitsuhiko: Should Developers Still Read AI-Generated Code Line by Line?

Redis creator antirez and mitsuhiko engaged in a public discussion about GPT 6 Astra's coding performance: antirez argued that when many excellent programmers get poor results from the model, it's not the model's fault — the skill set required for good programming has simply changed; mitsuhiko pushed back based on his own C language project experience, and the two sharply divided over whether one needs to read locally generated AI code. The exchange so far remains an exchange of views plus a suggested rewrite experiment, with no results yet.

Confirmed

• antirez observed that many excellent programmers report GPT 6 Astra works poorly for them, making him feel like he's "living in a parallel universe"; his explanation is that the skill set needed for good programming differs between past and present — there's some overlap, but it requires readjustment.
• mitsuhiko said he's currently in an adaptation period, using Astra to discuss problems rather than having it write large amounts of code directly.
• mitsuhiko reported concrete experiences: in a game side project, the C code Astra generated was often hard to read and frequently fell into "some goal-driven mode pursuing token efficiency."
• Comparing two models, mitsuhiko found that Opus 5.5 produced noticeably fewer "weird results" with almost no heavily compressed code.
• antirez's core claim: as long as the high-level design is expressed through prompts and validated, there's no need to read local code line by line; if local choices can't be validated by design, tests, or program goals, there's nothing to check — validation should stay at the design level.
• antirez suggested mitsuhiko run a "crazy experiment": spend two or three hours using only Astra to reimplement in C what he was originally going to do, to test its real performance.

Unconfirmed

• Neither party has reported whether the experiment was actually carried out or how Astra performed in a pure C rewrite scenario.
• The "changed skill set" explanation is antirez's personal view; mitsuhiko's complaints about code readability did not align with it, and the debate remains unresolved.

Why it matters

This exchange crystallizes a core divide in AI-assisted programming today: one side argues workflows should move up to high-level design validation and abandon line-by-line review of local code; the other insists that the readability and style of generated code directly affect engineering quality and maintenance costs. The first-hand comparisons from two veteran developers (antirez, author of Redis, and mitsuhiko, author of Flask) also offer a concrete reference for code-style differences between models like Astra and Opus 5.5.

2026-09-27 ~ 2026-09-27 · 12 related posts

• [source] antirez: Good programmers failing with GPT 6 Astra points to a different skill set — antirez · 2026-09-27
• antirez suggests a crazy experiment: build your whole project in C using only GPT 6 Astra — antirez · 2026-09-27
• mitsuhiko: Opus 5.5 rarely produces heavily compressed code unlike Astra — mitsuhiko · 2026-09-27
• mitsuhiko dislikes GPT 6 Astra's C code: hard to read and weirdly token-efficient — mitsuhiko · 2026-09-27
• [source] antirez challenges mitsuhiko: why read AI-generated code if the design is verified? — antirez · 2026-09-27
• antirez on AI code: if local choices can't be verified by design or tests, why check? — antirez · 2026-09-27
• [source] The thread starter: why read AI-generated code if the design is prompt-verified? — mitsuhiko · 2026-09-27
• antirez: Reading AI code is no longer feasible — he now trusts tests and QA instead — antirez · 2026-09-27
• mitsuhiko on Astra: expensive, narrow edits, and a frustrating experience to review — mitsuhiko · 2026-09-27
• antirez: RL will push AI-generated code further from what humans consider good — antirez · 2026-09-27
• antirez regrets not finding time to hand-write small poetry-style programs — antirez · 2026-09-27

1 near-duplicate retellings: mitsuhiko