Welcome back from Labor Day and a long summer. In today’s update: Building AI Review Loops, our upcoming Oktoberfest & more.
UPCOMING EVENTS
Beyond tonight’s Rooftop Social, here’s what we have coming up for you:
F&O x Wharton: Garden Social - Philly Special · Monday Sep 14, 6 PM · PHL
We’re excited to join the Wharton MBA Entrepreneurship Club for a Philly special. We’ll host a happy hour in center city followed by a dinner with Founders & Builders.
Founders Breakfast Club · Thursday Sep 18, 8-10 AM · West Village
Coffee, croissants, matcha, and co-working with your fellow founders & operators. Bring a demo. Ask for advice. Or just enjoy the company and space.
F&O Oktoberfest · Thursday Sep 24 6:30-8:30 PM · Watermark
It’s that time of the year again - we’ll swap our rooftop social for a waterside beer garden to honor the tradition of Oktoberfest. Join us for Brez’n & Beers at Watermark later this month.
BUILDING AI REVIEW LOOPS
Your AI says it’s done. But the output is sloppy. What now? What's worked really well for me: Building a review, fix, and recheck loop into my workflows.
- Give the reviewer clear instructions. Ask a separate review pass to check the agent’s work against the stated requirements. For code: bugs and missing error handling. For writing: unsupported claims and gaps in the argument. For front end: Is it clean, or are there obvious visual issues? You can also use another model (e.g. Claude to review ChatGPT) or a dedicated reviewer in an API workflow.
- Make it share the evidence to speed up human review. For frontend work, I require the agent to open the page, test interactions, and inspect desktop and mobile screenshots - and to share the screenshots with me. For spreadsheets, reconcile totals. For APIs, test the endpoint. “Done” should come with something you can inspect on the spot, so you can review the agent’s work right away without leaving the conversation.
- Close the loop. Have the AI fix valid findings and repeat the affected checks. Limit revision rounds and flag unresolved issues for human review.
Try adding something like this to your agent’s instructions:
Check the result against my requirements. Test the main workflow and inspect the output. If you find issues, fix them, then check again. Flag anything you can’t resolve or verify. Show evidence and share your review comments with me.
This hack has saved me a huge amount of time, headaches, and frustration. The model now catches many obvious issues itself. When it misses something, I can often spot it in the screenshot or analysis it shared in the chat and tell it what to fix, without opening the app or code base. Screenshots don’t replace functional testing, but they make visual review much faster.
Less time catching avoidable mistakes. More time doing the work you actually enjoy.
FROM THE COMMUNITY
Sharing a lovely reflection from Daria after joining us at our latest Rooftop Social:
We’re grateful when members share what Founders & Operators means to them. Follow our new LinkedIn page to stay connected, tag us in your posts, and keep up with upcoming gatherings.
Say hello, share your input, or drop us a note to be featured in a future edition. Ping us at [email protected].
See you soon,
Stephan & the F&O Team