SDD is BDD/TDD for the AI Era — A Guide for Software Crafters
The Software Crafters community has been arguing for engineering discipline since long before AI coding tools existed. Clean code. Continuous refactoring. Test-first. The boy scout rule.
Those values do not become less important when an AI is writing the code. They become more important.
The Core Analogy
You already know this principle: write the test first, then the code.
TDD forces you to articulate what correct behavior looks like before you implement anything. The test is a specification. The red-green-refactor loop is a verification cycle.
SDD applies the same principle to AI-assisted development:
Write the spec first. Then let the AI write the code.
Same discipline. Same intent. Adapted for the reality that the "developer" is now an AI agent.
Why This Audience Gets It Immediately
Software Crafters are already convinced that:
- Engineering discipline matters
- "Move fast and break things" accumulates debt
- Verification is not optional — it is part of the definition of done
- Shared understanding between team members prevents bugs
Planu does not need to convince you of any of this. It just shows you how to apply these principles to AI development.
The Parallel Is Exact
| Software Crafters Concept | SDD Equivalent |
|---|---|
| User Story + Acceptance Criteria | Spec + Acceptance Criteria |
| Test First | Spec First |
| Definition of Done | DoD gate in Planu |
| Code Review |
validate tool (automated drift detection) |
| Refactoring safely | Spec + validate = safe AI-driven refactors |
| Pair programming | Human + AI with shared spec |
The workflow is familiar. The tooling is new. The discipline is identical.
What Makes AI Development Different
There is one important difference from traditional TDD: the AI has no memory of your decisions.
When you pair program with a human colleague, they remember the discussion you had last week about error handling patterns. They remember why you chose that architecture. They can say "wait, that contradicts what we decided in the auth layer."
An AI agent starts fresh every session. Without a spec, it makes decisions that may contradict previous sessions. Over time, the codebase becomes inconsistent — not because the AI is bad at writing code, but because it lacks the context that a human colleague would carry naturally.
The spec is that persistent context. It is the shared understanding that survives session boundaries.
Drift Detection: The AI-Era Code Review
One of Planu's most useful features for Software Crafters is drift detection.
After you have an approved spec and an implementation, running validate checks whether the implementation still matches the spec. This is like a continuous code review that runs against an explicit contract, not just style conventions.
When the AI refactors something and accidentally breaks a behavior the spec defines, validate catches it. When a feature drifts because someone added "just one small thing" without updating the spec, validate catches it.
This is the "refactoring safely" pattern applied to AI development.
Collaboration Ideas for the Community
Conference Talk: "From Vibe Coding to Spec Driven Development"
A live demo comparing the two approaches. Format: 45-60 minutes with a hands-on section. This works well for any conference where developers are adopting AI tools and wondering how to maintain quality.
Blog Series: XP Meets AI
- "What XP Practitioners Need to Know About AI Development"
- "How to Apply TDD Discipline to AI-Assisted Development"
- "Why Your AI Code Reviews Are Failing (And How to Fix Them)"
Podcast Topics
- "The AI Quality Problem No One Is Talking About"
- Compare to early TDD adoption: same resistance, same eventual acceptance
- The parallels between TDD's critics in 2003 and SDD's critics in 2026
What Makes Planu Right for This Community
- No cloud required — data stays local, no privacy concerns
- Language agnostic — works with Python, Go, Java, Ruby, .NET, not just JavaScript
- Tool agnostic — Claude Code, Cursor, Windsurf, Zed, Cline, Continue
- Open standard — built on MCP, the open protocol for AI plugins
- Engineering discipline built in — the workflow enforces spec-first, not optional
The last point matters. Planu is not a suggestion. The workflow requires a spec before implementation. You cannot skip it without explicitly overriding it. That is a deliberate design choice — the same reason TDD frameworks make it easy to write tests first, not just possible.
One-Line Pitch
"Planu is what happens when Software Crafters discipline meets AI development."
Planu — Spec Driven Development for AI coding agents. Get started
Top comments (0)