Substrate engineering
summary
Claims that prompt engineering alone hits a ceiling — languages and frameworks set hard limits — and that rising LLM velocity worsens unnoticed bugs because better automation reduces human attention. Demonstrated failures include broken GitHub Actions YAML that only an editor error reveals, and proposals to replace ad hoc YAML/JSON schemas with typed, terminating configuration languages (example: Dhall compiles to JSON/YAML and enforces totality). Recommendation: invest in safer substrate, stronger type checks, and tooling that prevents common deployment hazards.
AI-written summary · about
chapters (7)
- 0:00Software engineering needs substrate engineering
- 3:57Prompt engineering not enough for LLMs
- 9:12Harder LLMs harder to review
- 12:04What is substrate engineering
- 13:29Tooling and configuration as substrates
- 21:21Safer configuration languages
- 23:08Responsible deployment and LLM future
related talks
50mFrom Prompt Hacking to Architectural Determinism: Engineering Reliable GenAI Systems
33mSkill Issue
33mCorrectness Too Cheap To Meter: Formal Verification and LLMs
38mPhishing LLMs: Reeling in the Machine
20mBuilding Deterministic Infrastructure for Non Deterministic AI Agents
47mHarnessing LLMs for Next-Level Security Analysis
32mThe Sorcerer’s Apprentice: The craft of building reliable solutions with LLMs
01h 00mAI Safety through Programming?