Fresh from the workshop: an AI agent running a SaaS that has a built-in AI tool

Fresh from the workshop: an AI agent managing a SaaS product that itself has a built-in AI tool. The label is probably "production pipeline with agent supervision."

How it works

Dr. Christina records videos on her phone. She chats with Leonie, her AI assistant, to say "I'm done" — and gets a link to upload the videos.

Leonie then runs the flow: uploads to the web app via API, handles basic editing, instructs the built-in AI tool for the rest of the editing (Leonie has a template), and notifies when the draft is ready for human review.

If something goes wrong, Leonie tries to handle it herself. If it's too technical, she contacts Nexus, our infrastructure AI agent. Or she contacts Christina, if there's a choice to be made.

Result: roughly 70–80% of end-to-end time eliminated, compared to doing the steps manually.

Lessons learned

1. Set rules (deterministically) where possible. Handle deviations with an AI agent.
2. 10% is "AI" and 90% is plumbing — handling large files, communicating status, trial-and-error when API docs say one thing and tests show another.
3. Bad ROI the first time, mostly due to the learning curve.

Share if you're building something.

Tech: Next.js, AWS S3, Postgres for status tracking, Descript API for video editing, Telegram for chat, plus agent infrastructure from before.