A Practitioner's Guide to AI-Assisted Coding

You shipped it in a weekend.
Can you fix it at 3 AM when you get a customer call?

AI writes the code now. The hard part is trusting what you can't fully read —
whether you're an engineer signing off on 2,000 lines a day, a founder
whose whole product was written by something else, or a student wondering how
much you're learning.

The mindset shift we need:
From being an AI-assisted Developer-> To AI-assisted System Architect

A practitioner's talk on building the harness: how to run agents hard without
ending up with a system you can't defend. A little bit of spec-driven development and context engineering -- what's the best tool for you ?

- If you never had to stop it, that's not a clean run — it's one nobody watched.
- Asking nicely doesn't always work. Constraints hold only when something checks them.
- Go only as fast as you can see. How big a task, how often you look, when to
stop trying — all the same call.
- Your corrections are an asset. The same fix you make by hand every other week should
become something the agent stops needing.
- What not to hand over. Speed is worthless if you're moving in the wrong
direction, and delegating the work you haven't learned yet has a cost that
shows up later.

Not a Cursor or Claude Code tutorial . Not a 10x promise. Bring the failure you couldn't explain and everyone will try to find a way to solve it.