Specs over Code: How AI is Changing Software

With AI coding assistants generating good-enough first drafts, implementation details and syntax mastery are becoming less important than specifications and systems design. This video explores where that shift is heading — and where it can go wrong.

The core idea: small, well-specified “black box” modules that AI agents can generate, test, and regenerate on demand, connected by clear contracts. But this architecture comes with real risks — duplication bugs across modules, operational complexity, code humans can no longer maintain without AI, and uncertain long-term AI costs.

Topics covered:

Why the focus is shifting from implementation mastery to specification and testing
Small “black box” modules built on contracts, TDD, BDD, and specification by example
Risk: duplication bugs spreading across modules
Risk: operational complexity — an extension of the microservices problem
Risk: architecture that isn’t fit for human modification
Risk: unclear long-term AI/token costs