We’re in the middle of another abstraction leap.
Just like compilers, cloud, and containers before it, AI-generated code started with hype, fear, and broken things—and might still end with progress.
But only if we build it right.
LLMs aren’t evil—they’re just monkeys with GPUs.
They’ll generate code with confidence, validate it with tests they wrote themselves (if any), and call it a job well done.
This talk introduces a working model for AI-assisted development that actually holds up under real-world use. It’s built on three ideas:
– The Chasm: the dangerous gap between what we meant and what we asked for.
– The Craft: the human skill to spot when AI gets it “technically right” but still wrong.
– The Chain: the Intent Integrity Chain, a structured flow of prompt → spec → test → code, where each output is validated externally—by humans or deterministic systems, never by the same model that generated it.

This isn’t a thought experiment or a product pitch. It’s a blueprint for building real software with AI, without sacrificing trust or intent.

Trust your knowledge.
Trust your tests.
Never trust a monkey.