AI coding assistants promise massive productivity boosts. In reality, most developers experience limited (or even negative) impact. What's up?
In this talk I give a couple of my heuristics and guard rails that make AI coding assistants worth your time.
* The smell test: the importance of experience
* The goldilocks zone: work sizing / parallellism sweet spot
* Guard rails: architecture, design, top-tier test automation
* Breadcrumbs: spec-driven but not too much. Clear acceptance criteria. Example of good code and tests.
* DON'T trust and verify