Generative AI Strategy

Chip Huyen just gave a talk with a title that will sound painfully familiar to anyone in a tech org right now: leadership needs us to do generative AI, what do we do? She delivered it at Fully Connected, and the whole thing grew out of conversations with friends who kept hitting the same wall. Everyone's been told to have a generative AI strategy. Almost nobody has been told what that strategy should contain.

Huyen's answer isn't a list of use cases or a vendor comparison chart. It's a framework for thinking through the problem before you start building anything, because most teams skip that step entirely and jump straight to a chatbot demo. The talk is explicitly a work in progress. She says several ideas are still being fleshed out, and she plans to turn it into a full written post once she has time, which tells you this is less a finished manifesto than a live snapshot of how one of the sharper voices in applied ML is currently reasoning about the problem.

What's notable is the format itself. Huyen crowdsourced feedback before finalizing the talk, posting earlier to ask what people wanted covered, and she credits Kyle Gallatin, Goku Mohandas, Han-chung Lee, and Jamie de Guerre for shaping the final version. That collaborative loop matters more than it sounds, because generative AI strategy right now is genuinely underdeveloped as a discipline. Most of the public discourse is either breathless hype or engineering-level tutorials on fine-tuning. There's very little in between that helps a team lead or a director figure out what generative AI is actually for inside their specific organization.

She's also released the slides separately, joking that she couldn't get them centered on the page, a small detail that undercuts any sense that this is a polished product pitch. It reads more like someone thinking out loud in public and inviting correction, which is a rarer posture than it should be in a field where everyone claims to have the playbook already figured out.