Navigating AI Native Development
summary
Argues software shifts from code-centric to spec-centric development, demonstrates Bolt.new, Cursor and Copilot generating working UIs (a Bolt app shows 242 votes), and builds a two‑week prototype called Specify that uses GPT‑4o to turn a brief into a spec, tests and code (dice-roll module: initial CLI run shows 2 pass/9 fail then respin to 7 pass; earlier pluralize demo hits ~64% passing). Notes LLM limits (context window, token granularity, poor diffing) and stresses code remains source of truth until specs are stable and spec becomes long-lived guiding regen.
AI-written summary · about
chapters (7)
- 0:00AI native development intro
- 2:27Demo: Bolt.new
- 12:31Prompt engineering at Tesla
- 19:10Code-centric to spec-centric
- 30:35Demo: Specify
- 43:19Future of AI-native dev
- 45:45Q&A
related talks
54mKeynote: AI-Powered App Development
37mSpec Driven Development: AI Assisted Coding With Doc at the Core
50mSpec-driven Development using Coding Agents
29mSpecs Don't Hallucinate: Deterministic Code in the Age of AI
15mThe AI-Augmented Engineer [...]
14mThe 4 Patterns of AI Native Development
12mNavigating the World of AI Development Without Going Insane
17mAI, You complete me