Why I'm building TrueBDD in the open

Building a tool whose whole point is to make the code safe to throw away and regenerate from the spec.

AI generates code faster than any of us, but as long as code stays the source of truth, nothing fundamental changes: engineers still review everything, and every link in the

product manager → engineer → AI chain

loses information.

Specs as the source of truth collapse that chain: the spec is meant to be the artifact everyone works on, so there are fewer translation steps and less gets lost between intent and what ships.

That's why I'm building TrueBDD.

It's a CLI, built on one bet: the spec is the source of truth, and the code is something you regenerate. The work is now in the open.

The idea

a Gherkin-grade behavioural spec plus an explicit architectural contract carry enough for an AI to rebuild the code from scratch. What survives a rebuild is the observable behaviour, not the byte-for-byte shape. Two rebuilds from the same spec won't produce identical code — the AI can invent endpoints and internal paths to satisfy the spec, as long as the data models and exposed endpoints the architecture pins down stay intact.

Why this is hard

Pre-AI, code was inevitably the only authoritative artifact — you discover detail mid-implementation, work around third-party limits, absorb UAT feedback, and the spec quietly diverges. Coding agents don't fix that; agents drift too, and a stakeholder can still ask for a checkbox where a dropdown was specified. So the code stays the only honest record of what the system does. The point of TrueBDD is to move that record up to the spec.

How you'd know it works

delete all the code, regenerate it from the spec, and check whether the new build still passes every behavioural assertion the spec carries. If it does, the rebuild is correct, regardless of how the code came out different. That's the whole contract — behaviour preserved under regeneration, not byte-identical regeneration.

I want to be straight about the stage.

Today the tool runs one level down from that: the spec drives the workflow, the code is still hand-held, and full regeneration is ahead of me. It's a prototype that already does real work, not something I'd hand to your team yet.

Repo's open. 🔗 https://lnkd.in/dNXNtWvU 🔗

If you build in this space, I'd like to know your thoughts!