"My job is to write loops": the coding shift Armin Ronacher resents but expects
Boris Cherny, who built Claude Code, says he no longer prompts Claude — he writes loops that prompt it for him. Armin Ronacher thinks that future is coming whether he likes it, and he does not.
Ronacher splits the agent loop you supervise inside a session from the harness loop that runs outside it, deciding on its own whether to continue. His worry isn't speed: LLMs won't make illegal states unrepresentable, piling defensive checks on errors that can't happen.
The thread turned it into a labor question. Engineers describe two camps — agents-go-brrr versus the people drowning in their PRs — and a darker read: the Claude Code creator's job is to move tokens, like a brewery CEO's is to move kegs.
Two kinds of loop, and the one that scares him

Armin Ronacher's essay separates two things people both call "the loop." The agent loop runs inside a session: the model calls tools, reads results, and decides it's done. The harness loop runs outside it — work enters a queue, machines attempt it, and an external system decides whether to continue, spin up a fresh session, or route the task elsewhere. He quotes Boris Cherny, who built Claude Code: "I don't prompt Claude anymore. I have loops running that prompt Claude and figuring out what to do. My job is to write loops."
Ronacher's reservation isn't throughput. It's that models produce "code that is too defensive, too complex, too local in its reasoning": they add fallbacks instead of making bad states unrepresentable, and, in Karpathy's phrase he borrows, are "mortally terrified of exceptions," guarding against errors that the type system already rules out. Loops amplify that, layering defensive cruft with each pass. He grants where loops shine, mechanical work like porting, citing the reported move of parts of Bun from Zig to Rust and his own MiniJinja port, and he states his own line plainly: "I want to understand the code I ship." The unsettling section is titled You Cannot Quite Opt Out: even a maintainer who refuses loops can't escape them, because attackers and security researchers already run loops against his code.
The null-check tell, confirmed in the wild
The complaint about defensive slop drew engineers who fight it in review daily. One who shares Ronacher's "make illegal states unrepresentable" instinct explained why models resist it:
"Most programmers still think in terms of picking up pieces and fixing errors at the point where the error message pops out rather than making it so the error can't happen… AI loves to sneak past a '.JustSetItAndIgnoreAllThePreAndPostConditions(string)' method." — jerf
Another named the exact code smell and asked the labs to fix it at the source:
"In python, it often comes as hasattr checks on types that are defined to have that attribute, in a code base that is fully type-checked. Why do they do that? Is it from pre-training or re-enforcement? If the latter, can the labs please fix this?" — a commenter on LLM defensive-coding patterns
"His job is to move tokens"
The thread's strongest move was to treat the loop less as a technique than as a workplace reorganization. One engineer described the split it's already creating:
"We have two camps of people at work, the first camp are the agent goes brrr. They don't understand the code they write… The second camp is people who are inundated with PRs, are holding the line on quality, and just exhausted." — a commenter describing two camps at his job
That framing produced the sharpest line about why the Claude Code creator advocates loops at all:
"If the nation's drinking skyrocketed, we wouldn't be sitting here wondering why the CEO of Budweiser isn't advocating for temperance. His job is to move kegs, just like Boris' job is to move tokens." — a commenter on the incentives behind loop advocacy
Not everyone read Ronacher as a hype-man. One commenter heard the opposite — resignation, not evangelism:
"I read it as the exhausted sigh of resignation from someone who sees where the wind is blowing whether they like it or not… you can certainly opt out, but you won't be opting out of AI, you'll be opting out of the industry." — a commenter on the essay's tone
There was a quieter, economic counter to the whole panic, too: looping only pencils out while tokens are cheap. "If token costs are nil, then you can afford to run verification and generation through the same models. If token costs are high, then you will go broke verifying code sprawl" — a commenter tying loops to token subsidies. That is the thread's real punchline against the essay's fatalism: the loop isn't an inevitable law of software so much as a feature of subsidized compute. Take away the subsidy — the same one being litigated one story over — and "my job is to write loops" may turn out to have been a phase, not a future.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Anthropic locked away Mythos as too dangerous. A hobbyist benchmark says: maybe
Mythos and its filtered twin Fable were yanked offline before anyone outside could measure them. So one developer rebuilt the test from the nine bugs Anthropic said Mythos found, pointed a dozen other models at the same files, and let them hunt blind.