Building with agents at a startup

what this covers

• What does a founding engineer do at an AI startup?
• How do you ship LLM-powered features without over-engineering?

Most of my week is not prompting models. It is deciding what should be automated, what needs a human in the loop, and how to keep the product understandable when the stack gets weird.

Shipping beats spectacle

Early teams win when the loop is short: talk to users, ship a thin slice, measure, repeat. Agents are useful when they remove repetitive glue work — not when they replace judgment.

What I am optimizing for

• Reliability over demo magic
• Clear interfaces between product, data, and models
• Writing so the team (and future me) remembers why we chose a path

If you are building in this space, say hello.