Ball: drop pull requests. Julius: 400 perf PRs in 30h
Thorsten Ball spent the week arguing nobody needs pull requests anymore and published the SOC2 pipeline he pushes straight to main with. Meanwhile Julius let an agent merge almost 400 performance PRs in 30 hours, and Theo says you can feel it.
The big disagreement
Thorsten Ball says pull requests are over, and posts his SOC2 pipeline
Ball ended the week with receipts rather than a slogan: he published the push-to-main pipeline his SOC2-compliant team runs, where Buildkite tests and builds, GitHub Actions runs security checks, a failed check blocks deploy and pings the commit author in Slack. His original argument was that Opus 5.5, Fable or Astra can review four times each, so humans reviewing diffs adds nothing.
Sure sounds like a lot of people don't trust their colleagues to work like this was their own project.
Receipts
Julius's agent merged almost 400 perf PRs in 30 hours, Theo felt it
Julius says an agent spent 30 hours merging close to 400 pull requests aimed at web performance, reporting 86% fewer renders when another thread streams, up to 35% less JS CPU and a roughly 10% smaller heap. Theo, who runs T3 Code, says the latest nightly's performance is insane and almost feels fake. Mobile is next, according to Julius.
T3 Code already felt fast but the latest nightly's performance is insane. Almost feels fake, like the data is hard coded or something.
Opus found a leaked .env behind a $90,000 bill, Theo found Gemini 4 Argon wrong over half the time
Two reliability reports landed in the same week. @justlingonberry says Opus 5.5 traced a $90,000 Google bill to Codex publishing their .env to Cloudflare Pages, while Codex found nothing and flagged the chat as a cybersecurity risk.
This model is a straight up liability. Keep it away from your code.
How they actually work
Pocock blocks any agent-authored file over 5,000 tokens at commit
Matt Pocock's prompt of the week asks an agent to write a pre-commit hook rejecting files over 5,000 tokens, counting tokens as bytes divided by four, running only on staged files, skipping lockfiles and codegen, and printing explicit splitting instructions in the error. He calls it the simplest investment he ever made in his codebase.
Model and agent watch
Theo puts Muse Wonder level with Opus 5.5, and Sol a tier below
Theo's current rankings put Opus 5.5 and Muse Wonder in S, GPT-6.1 Sol in A, and everything else below a gap, with the caveat that he likes them a similar amount rather than rating them equally capable.
Muse Wonder Full found a small PR someone else had filed and said "you should probably merge this instead
Open tabs
dax says his own DeepSeek inference rack beats every vendor on price/perf
dax reports a first test run on a new GPU rack with their DeepSeek setup, beating every vendor they found on price/performance.
Anthropic replaced git after merges went from 2,500 to 30,000 a day
Anthropic says it migrated its monorepo off git to an in-house version control system in five months, after daily merges grew from around 2,500 in May to over 30,000 in September.
Frequently asked questions
What did Thorsten Ball publish instead of doing pull request reviews?
Ball published his team's push-to-main pipeline that uses Buildkite for testing and building, GitHub Actions for security checks, and automatically blocks deployments and alerts the commit author in Slack if a check fails.
How much can an AI agent improve web performance?
Julius reported an agent merged close to 400 performance pull requests in 30 hours, resulting in 86% fewer renders when another thread streams, up to 35% less JavaScript CPU usage, and roughly 10% smaller heap size.
What security issue did Opus find that Codex missed?
Opus traced a $90,000 Google bill to Codex accidentally publishing their .env file to Cloudflare Pages, while Codex had found nothing and flagged the chat as a cybersecurity risk instead.
What is a pre-commit hook and why did Matt Pocock add one?
A pre-commit hook is a script that runs before code is committed. Pocock created one that blocks any agent-authored file over 5,000 tokens (tokens counted as bytes divided by four) to prevent files from growing too large to review.
Why did Anthropic replace git with a custom system?
Anthropic migrated off git because daily merges in their monorepo grew from around 2,500 in May to over 30,000 in September, requiring a custom in-house version control system to handle the volume.