The Agentic Workflow

Spec-driven development means writing a detailed specification before letting an agent write code, giving the agent explicit success criteria, constraints, edge cases, and architectural context instead of a vague natural language request. The gap between "build me a login page" and a structured spec that defines auth flow, error states, styling conventions, and API contracts is the gap between throwaway code and production-ready output.

Test-Driven Agentic Development (TDAD) adapts the classic test-driven development cycle for agent-assisted coding: you write tests first and then direct the agent to implement until all tests pass, using the test suite as executable specifications that define correctness rather than leaving that definition to natural language. This inverts the typical agent interaction, where you describe what you want and hope the agent interprets it correctly; instead, the tests encode your requirements unambiguously and the agent iterates until they pass.

Eval-driven development treats evaluations as first-class development artifacts, measuring agent behavior against defined criteria before, during, and after every change, analogous to test-driven development but designed for non-deterministic AI systems. Instead of manually checking "does this seem right?", eval-driven teams build evaluation datasets that encode expected behavior and run them automatically whenever prompts, tools, or models change.