Scheduled Claude Code Routines Shipped 1,531 PRs to a 17-Year-Old SaaS in a Month
In about a month, scheduled Claude Code routines shipped 1,531 PRs to Preceden, a 17-year-old SaaS, across 20 batched deploys with none reverted. The owner says MRR impact isn't visible yet.
Who this is about: Solo-owned SaaS (Preceden, a timeline maker)
What it takes
- Cost
- From $20/monthClaude Code: Included with Claude Pro ($20 a month billed monthly), Max, Team and Enterprise plans, but not the Free plan. You can also pay for API usage instead.Price checked Oct 24, 2026 on claude.com. Prices change; check before you buy.
- Setup time
- A few weeksOur estimate from the source, for one business.
- Skill needed
- Developer skillsYou build or run your own code, even if AI writes it.
Explain it simply
A founder who no longer has time for his old software product set up AI coding agents that run on a schedule on his Mac. They fix bugs, add tests, improve help articles and tune pages for search, each as a separate change that is reviewed and tested before release. When they need a decision, he gets one message a day.
What was done
Several scheduled routines run around the clock in the Claude Code desktop app, each a fresh session with a written playbook: an hourly SEO agent, an hourly code agent, a nightly dependency agent, a daily listings agent, a rank tracker and a weekly retro agent.
Guardrails: no billing, auth or email logic, no schema changes except indexes, size caps per change, and a kill-switch file that halts everything.
Every change is its own pull request into a dev branch. Twice each weekday a deploy agent merges to master after the full test suite, a Claude security review and a second-opinion review from the Codex CLI.
The agents have API access to GitHub, read-only production Postgres, logs, Sentry, Search Console, Help Scout (read-only) and the deploy service.
Decisions come to him as "asks" in one 9am Slack message, and his answers become standing rules every agent reads first.
How it went
He reports 1,531 PRs in a month, mostly support articles, tests and bug fixes, and 20 batched deploys of 5 to 275 changes each, none reverted. He says there has been no downtime, no security issues and no bug reports caused by the changes.
- The work was deliberately limited in scope while he learned how to build the system around it.
- Revenue impact is unknown so far; he expects to see it in MRR in the coming months.
- Anything touching production data becomes a script he runs himself.
Prompt
Set up a daily routine for [repository]. Each run:
1. Pick one small improvement to the codebase or SEO (a bug fix with a failing-then-passing test, a missing test, a clearer help article, a page fix).
2. Make it on its own branch and open a pull request into [dev branch]. Never push to [main branch].
3. Don't touch billing, authentication, email sending or database schemas.
4. Write a short note: what you changed, why, and how you checked it.
If you need a decision, list it as a question with options and your recommendation, and stop.Adapted from his suggested way to start small.
Setup notes
- Start with one daily routine that opens one small pull request, and review every one by hand at first.
- Protect the main branch and require tests to pass before anything merges.
- Keep a kill-switch file and a list of areas the agents may never touch.
Business fit
Who this helps, and how:
Owners of older side-project SaaS: keep a product improving when you no longer have time to work on it.
Solo founders- Startupsclear a backlog of small bugs, tests and help-article fixes while the team builds features.
Related
- Evidence grade DA 112-Member Gym Swapped a $3,600 SaaS for Its Own App With an AI CoachLight evidence, using Claude Code, Claude and Expo
- Evidence grade DA Startup's Grok Bot Signed Up for Competitors' Products and Found Where Users Would Drop OffLight evidence, using Grok Bot
- Evidence grade BShip a product in an afternoon with an AI code editor, then sell sponsor slotsStrong evidence, using Cursor
- AI for productDepartment hub: AI for product, 4 examples
Similar tools
Other tools for this job, and how they differ. Chosen by our editors, never paid for.
- OpenClaw
Self-hosted framework for long-running autonomous agents; full control, but setup, upkeep and security are yours.
- Grok Bot
Bots run routines on their own cloud computer with a terminal and can hand coding work to the Claude Code and Codex CLIs.
Stack: Grok Bot + Claude Code and Codex CLIs: route each task to a model
“Strong evidence” (grade B) means: The person who did the work describes the setup and results in detail. How we grade. Spotted a mistake? Suggest a correction.