Never Trust a Monkey: The Chasm, the Craft & the Chain of AI-Assisted Code
summary
Models get better but developers trust them less: surveys show ~76% see hallucinations, 96% distrust AI, 42% still commit AI code to production while only 48% always check it. The proposed workflow uses specdriven development and BDD to generate deterministic tests from human-verified specs, hashes and locks only assertions (hash assertions) so LLMs must pass tests, and adds context engineering to avoid endless retries. A demoed Intent Integrity Kit scaffolds the chain and claims ~4x improvement when product context is provided.
AI-written summary · about
chapters (10)
- 0:00The AI trust problem
- 2:06Analyzing AI code quality
- 7:20Why software development fails
- 12:24AI's impact on code reliability
- 17:17The limitations of code reviews
- 19:35Introduction to spec-driven dev
- 24:01Verifying the intent chain
- 31:02Building an integrity chain
- 37:02Context management and tools
- 40:50Demo and final takeaway
related talks
51mNever Trust a Monkey: The Chasm, the Craft, and the Chain of AI-Assisted Code
32mCan We Trust AI-Generated Code?
54mNever Trust a Monkey: from AI Gobbledygook to Deployable Code
26mThe Trust Problem: Designing Quality Frameworks for AI-Generated Code
29mBuilding Trust in AI Agents
33mShipping Fast, Breaking Silently: Why AI-Generated Code Needs a Second Opinion
26mAl and Software: The Good, The Bad, The Scary
29mWhy did I break production? Is AI the sickness or the cure?