For months, I worked with Claude Code the same way: I'd describe roughly what I wanted, the agent would code, I'd correct, the agent would recode — and after a few hours we'd drifted far from the original problem without either of us noticing. The code worked. But it no longer matched what I'd had in mind at the start.
That's not an AI problem. That's a context problem.
BMAD — Breakthrough Method for Agile AI-Driven Development — is the open-source framework that changed my approach. The central idea: build context before writing a single line of code. Not as a constraint, but as the thing that separates a project that holds together from one that accumulates technical debt from day one.
AI agents are extraordinarily good at executing. They're much less good at guessing your intentions. Give them a vague brief and they'll produce something plausible — not necessarily what you need.
The real cost of that vagueness doesn't show up in the first session. It shows up three weeks later, when you add a feature that doesn't integrate cleanly, when you realise the architecture can't support your next requirement, or when someone else joins the project and there's no way to reconstruct why a given decision was made.
BMAD forces you to answer those questions before you start coding.
The framework is built around a sequence of three specialised agents, each producing a structured deliverable that feeds the next:
First, you clarify the why. What problem are you actually solving? For whom? What are the real constraints? What's in scope — and crucially, what isn't? This is the deep interview phase, where no ambiguity is left hanging. The Analyst produces a 1-2 page brief that acts as a compass for everything that follows.
The brief becomes a Product Requirements Document: Epics (major workstreams) broken down into Stories (isolated, bounded tasks). This breakdown is the core of the method. A well-written Story fits into a single working session with an agent. It has a clear acceptance criterion. The agent can't go off-track because the scope is defined.
This is where technical decisions get locked in: stack, file structure, patterns, API choices. For Nortrip Toolbox, this is where we documented why we chose Supabase over Prisma, or how API routes are structured. Three months later, those decisions are traceable.
A fourth role — the PO — validates coherence across the three documents before development agents are launched. They also shard the documents into files the dev agents can actually use: each agent receives only what it needs.
I've applied BMAD on two recent projects: an accounting reconciliation feature for Nortrip Toolbox, and a quote management system for a craftsman client in Gironde.
The most immediate difference: recovery speed when something goes wrong. When an agent heads in the wrong direction, I have the documents to course-correct in a single sentence — no need to re-explain the full context from scratch. The Project Brief is there. The Story is there. The architecture is there.
The second difference: decision traceability. In six months, when I come back to this code — or when someone else works on it — the BMAD documents explain the why. It's not just code. It's project knowledge.
One important thing: BMAD isn't a heavy process you apply to everything. The framework calibrates to complexity:
That's what makes it usable day to day. It doesn't slow down your workflow — it structures it at the right depth.
If you're using Claude Code, Cursor, or any AI agent to build systems — automations, web apps, internal tools — and you've ever felt like you were going in circles without making real progress, BMAD is for you.
It's not a method reserved for tech teams. I apply the principles with SMB clients who've never written a line of code, purely to frame their requirements before we start building together. The result: fewer back-and-forths, more predictable delivery, and systems that hold up over time.
The framework is open-source and available on GitHub. But if you want to see how to apply it concretely to your project — whether that's a Make automation, a custom web app, or an AI agent system — we can work through it together.
I offer a free one-hour audit to scope your project using the BMAD method and identify the first Stories to build.
Book your slot here — and stop going in circles.
Benjamin Bertigny — Founder NBN IA. I build AI systems for SMBs and craftsmen. Based in Bordeaux, serving all of France.
Ready to automate your business?
Free 1-hour audit — we identify your 3 most valuable automation opportunities together.
Book free audit