Fast Models Need Slow Developers
summary
Codex Spark generates code at 1,200 tokens per second, about 20× faster than Sonnet/Opus at 40–60 t/s, so habits built around slow generation produce scaled technical debt. Validation and linting run continuously and instantly, cherry picking becomes practical — e.g., 75 component variations created by five sub-agents — and a persistent four‑file external memory system (agents.md plan.md progress.md verify.md) keeps sessions resuming instead of restarting.
AI-written summary · about
chapters (11)
- 0:00Introduction to the impact of fast AI code generation
- 2:29Historical context of model speeds
- 3:10Why AI inference speeds are increasing (Hardware/Stack optimization)
- 7:05The current developer landscape and risks of "slob"
- 8:27Playbook: Orchestrating models and sub-agents
- 9:56Playbook: Validation and automated testing
- 10:47Playbook: Cherrypicking and variety in output
- 12:07Playbook: Adopting a real-time collaborative mental model
- 12:53Playbook: Avoiding "slob" and active steering
- 13:54Playbook: Continuous refactoring
- 14:30Playbook: Context management and external memory systems
related talks
21mBuilding at the Speed of Thought
48mHow to Verify AI-Generated Code with Codex
21mDesigning in the age of the 10x engineer
15mDevelop at Idea Velocity
37mAI-Driven Developer Productivity at Scale
01h 00m10 tips to level up your ai-assisted coding
31mAI Coding Agents Don't Need Bigger Models. They Need Better Tools.
36mCool kids' way of generating code