Hand a small internal-tool annoyance to an AI coding agent, then turn its focused fix into a pull request without pulling an engineer away from larger work.
Source episode · 22:00
How I AI: Gumroad CEO Sahil Lavingia’s Workflow for Building Features 40x Fasterwith Claire Vo
22:00 to 25:00: Sahil Lavingia asks Devin to clean up Gumroad’s weekly Slack recap, then inspects the code and routes the pull request to a teammate.
A narrow pull request that removes the annoyance, updates any related prompt or logic, and includes enough evidence for a teammate to finish the merge.
AI software engineer by Cognition Labs
Step by step
Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.
5 steps
Pick a recurring annoyance with a crisp before-and-after. Sahil uses a weekly Slack recap that shows empty projects and treats backend-only work as a shipment.
In our weekly Slack recap, empty project sections appear even when nothing shipped, and backend-only work is described as a shipment. The output should show only projects with user-facing shipments. Keep the existing format for qualifying projects.
State the behavior change in plain language and include one real example. Mention any likely prompt, filter, template, or scheduled job the agent should trace.
Find the code that generates the weekly Slack recap and the AI prompt it uses. Make two changes: hide project sections with no shipments, and tell the model to summarize only user-facing features, improvements, and bug fixes. Do not include backend-only work. Use [example project] as the before-and-after case.
Assign the task in the correct repository with permission to inspect the generation path. Ask for a focused pull request rather than prescribing the exact file or implementation.
Work in [repository]. Trace the weekly Slack recap from data selection through the model prompt and final formatting. Implement the two behavior changes above, add or update focused tests, and open a pull request. Avoid unrelated refactors.
Let the agent locate the recap code, its prompt, and the filtering logic. Require a sample output so the result can be judged as a report, not only as a diff.
Run the focused tests and generate a sample recap containing one shipped project, one empty project, and one backend-only change. Include the sample and a short explanation of the filter and prompt changes in the pull request.
Read the sample output and the small part of the diff that determines inclusion. Route the pull request to the teammate who owns the tool when they can confirm repository-specific behavior or missing tests.
Address these pull request findings: [list findings]. Keep qualifying project sections unchanged, add any missing regression case, and update the sample output before requesting another review.
Turn an idea into a PRD, user stories, and a plan.
After the steps
How to recover when the loop fails and where human judgment helps.
Recover
Keep building
Log your workouts effortlessly without ever opening a spreadsheet. This workflow uses the Claude mobile app to parse a simple voice note and automatically update your workout tracking spreadsheet in the background.
Create a custom email management interface within Claude to read, triage, and draft replies without opening the Gmail tab. This workflow helps you reclaim focus and compound your learnings by training your AI on every interaction.
Build a fully automated system in Claude to generate personalized, interactive HTML proposals and client onboarding materials, saving hours on administrative work and creating a high-touch experience.
Join 100,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.