Course Wiki

Spec-Driven Development

Write the specification before the code: turning a vague idea into a spec and a task backlog an agent can execute.

Module 1’s cautionary example is a CLI built from one vague sentence — the agent happily produces something, and it is the wrong something. Spec-driven development inverts the order: describe expected system behavior first (users, interactions, interfaces, edge cases), turn the spec into a groomed backlog of focused tasks, and only then let the agent implement task by task.

The course walks a real project through the arc — from one-line idea to specification to backlog to working Django app — and shows that the spec is what makes agent output reviewable: with expected behavior written down, a human (or a QA agent) can verify the implementation against it.

Taught in

• Machine Learning Zoomcamp

  • Module 5: Deploying Machine Learning Models
    • Web services: introduction to Flask

• AI Dev Tools Zoomcamp

  • Module 1: AI-Native Developer Workflow
    • AI-Native Developer Workflow ## Related concepts

• Context Engineering

• Loop and Graph Engineering

• Coding Agents

• OpenAPI Contract