Becoming an AI engineering company
summary
Vercel describes a concrete path: wrap large models with retrieval-augmented generation using your proprietary search, automate checks with evals as unit tests, and expose domain logic as callable JavaScript/React components so the LLM can drive real UI. A prototype took about three months to become a usable product; initial runs cost roughly $0.66 per generation and felt slow but workable. The stack uses Next.js/React (Vue support exists), swaps expensive frontier models for middle tiers plus prompt engineering, and the demo actually runs.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 4:27AI disruption at Vercel
- 5:52AI engineering revolution
- 7:55From research to AI development
- 10:24AI engineering fundamentals
- 15:23Evaluating AI with unit tests
- 18:16Optimizing AI with feedback
- 20:45Introducing AI SDK
- 22:53Connecting AI to React
- 25:44Chatbot wrappers underrated
- 27:58Live demo of VZero
- 31:59Q&A
related talks
25mBuilding high performance, AI-native engineering teams
27mAI-Generated React Server Components
23mBecoming an AI engineering company
09mAI for React Developers: Opportunities, Learning, and Innovation
31mYou are an AI engineer
48mGoing Beyond Prompt Engineering
48m(Unpacking) AI's Impact on System Architecture
28mSoftware 3.0 and The Emerging AI Developer Landscape