Building the systems that build the software
summary
Claim: engineering shifts from writing code to designing sociotechnical systems that orchestrate AI code producers. Uses: StoryMachine experience, demos with Claude, Copilot/Cloud Code, Opus, Gemini, GPT‑5, BM25 search, vector DBs and RAG. Shows a toy system and live simulation where fast agent speed plus coarse tasks produces rising hidden debt and oscillating velocity; breaking work into checkpoints reduces debt. Benchmarks reveal SWE‑Bench misleading (Opus tops, Django‑heavy problems, no error bars). Skills emphasized: systems thinking, environment design, empirical rigor.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:03Software engineers' new role
- 3:11Complexity in AI systems
- 7:26Code review feedback loops
- 9:33Technical debt with AI agents
- 11:59Live demo: AI-driven systems
- 17:18Agent-friendly environment design
- 18:44Case study: AI customer support
- 22:34Evaluating AI systems rigorously
- 25:45AI benchmarks pitfalls
- 27:31Summary
- 28:15Q&A
related talks
20mBuild Systems, Not Code
25mThe role of software design in an AI world
56mFundamentals of Software Engineering In the age of AI
42mSoftware Tomorrow
33mBuilding an Agentic Design System
01h 09mThe Future of Software Engineering
27mThe new gravity of software engineering in the AI-first era
32mBuilding an AI Operating System