Dev let Claude Code run autonomously on his 17-year-old SaaS: 1,531 PRs merged in a month
mhmazur · x · 2026-09-21
Indie developer Matt Mazur let Claude Code run autonomously for a month on Preceden, his 17-year-old Rails timeline-maker SaaS, and it merged 1,531 PRs: 437 support-article improvements, 392 test-coverage additions, 162 Ruby bug fixes, 144 dead-code removals/refactors, 113 JS fixes, 98 marketing/landing-page improvements, 66 UI/copy fixes, 45 SEO fixes, 30 accessibility fixes, 16 performance fixes, and more.
No fancy harness — everything runs in the Claude Code desktop app with macOS caffeinate keeping the Mac awake, via scheduled routines (each a fresh session with a written playbook):
- SEO agent (hourly): docs, structured data, sitemaps, max 2 changes per run
- Code agent (hourly, offset 30 min): bug fixes require a failing-then-passing test as proof
- Dependency agent (nightly): one gem/npm bump per night, security alerts first, only releases ≥14 days old
- Listings agent (daily): measures AI-generated timeline pages' search performance, mines 20 new topics
- Daily rank tracker + weekly retro agent auditing the others
- Guardrails: no billing/auth/email logic, no schema changes except indexes, size caps, a kill-switch file
- Every change is its own branch→PR→dev branch, never master; mandatory security review before each merge halts the system on findings
- Twice daily a deploy agent merges dev→master: full test suite, Claude security review, independent Codex second opinion, CI — then deploys via Hatchbox API and watches logs/Sentry for 30-45 min with whole-batch revert capability. 20 batched deploys so far (5-275 changes each), zero reverts
- All tracked in plain files: 1,000 ledger entries, an idea bank, read-back checks
More from coding & agent
- User automates entire video clipping workflow with Nous Research's Hermes agent — Teknium · 2026-11-12
- As models improve, writing good evals gets harder and more important — amankhan · 2026-11-12
- Prime agent ships with day-one multi-agent communication, runs in Minecraft unmodified — sarahcat21 · 2026-11-12
- Autonomous agent edits just 1,024 weights to lift Qwen3.5 0.8B from 7.4% to 37.4% on MATH-500 — VictorSanh · 2026-11-12
- AI game jam with Anthropic, Google, OpenAI teams: cash prizes plus $50K continuation grants — chrisfirst · 2026-11-12
- Free AI game jam in SF Nov 14–15 with teams from Anthropic, Google, Meshy, OpenAI — chrisfirst · 2026-11-12