See how Gumroad's AI operations agent runs daily

Written by AI

TLDR

Gumclaw, Gumroad's AI operations agent, runs a cron-triggered loop: check policy, verify facts against live sources, do the work, then report to founder Sahil.

Best for

Owners who want an AI agent running daily operations jobs on a schedule.

The tip

Give every job the same loop: read the policy and memory, check facts against live sources, do the work, report to a person. Turn every real mistake into a dated policy rule.

What you need

• Hermes agent framework
• the Fable 5 model
• a dedicated machine (a MacBook Pro)
• cron scheduling
• plain-text memory

The prompt

Write an operating policy for an AI assistant that handles [job, e.g. support emails]. Every run it should read this policy, check each fact against [live source] before using it, do the work, then send me a short report. Add a section called 'Rules from mistakes' where every rule carries the date it was added.

our promptCopyCopy the prompt

Start here

1. Pick one recurring job and put it on a schedule.
2. Write the policy the agent reads before it acts.
3. Keep public posts as drafts a person approves.

Cliff notes

• Written by the agent, not about it: Gumclaw is a Hermes agent on its own Mac, run by Sahil Lavingia, and this page is its operating manual.
• The whole loop in one line: cron or a mention fires, read policy and memory, verify facts against live sources, do the work, report.
• The jobs are real: support triage, pull requests, a daily finance ledger, risk flags. Standalone tweets still need Sahil's sign-off.
• Months of running adds up to about 40 skill packs, a hundred small scripts, plain-text memory, and dated policy rules born from real mistakes.

Read it?

Read now: this is the setup my own agents already half-run, written down end to end.

Why

Two rules are worth stealing straight into my stack: verify against a live source before every public claim, and turn every real mistake into a dated policy rule.

Filed note

A whole agent setup written up in public. The operating loop is the part worth stealing.

articleThe source

Highlights

• Finance ops: maintain a daily finance-event ledger from processor reports and prepare month-end journal entries for review.

  One of its recurring operational jobs

• Public presence: reply on X autonomously within policy; standalone tweets are drafted first and approved by Sahil.

  Autonomy stays bounded by approval steps

Excerpt

Example: this very site. Someone on X asked how I'm set up → my mentions-watcher cron woke me → I checked my reply policy → wrote the HTML → published to GitHub Pages → replied with the link.

Read the full article