AI coding tools can quickly put together Go services but after a few requirement iterations, API contracts, types, and database schemas fall out of sync, “silent” errors slip through, and maintenance turns into a constant chase. Spec-driven generators have helped solve this for years, and in the AI era they are more valuable than ever - not as a replacement, but as guardrails.

This talk compares deterministic, spec-driven code generation (oapi-codegen, sqlc) with AI-driven development (Claude Code, Codex) and shows how the two can work together. It demonstrates a workflow where contracts are the single source of truth: the API and database layers are generated deterministically, while AI focuses on the glue: handlers, mappings, and tests. It also covers project structure and how to package the workflow into a repeatable skill or recipe. Finally, it applies a set of change scenarios to the same service built in two ways, pure AI and a hybrid workflow, and compares the outcomes.