A governed, structured, self-improving system for figuring out where AI agents belong across your operations — and making that real. Open, evolving, and freely available.
Every organisation hits the same wall. Not "should we use AI?" — but "how do we actually do this properly?"
Scattered experiments that don't connect, don't compound, and don't survive the person who built them.
Boundaries bolted on after something goes wrong, instead of being built in from day one.
Deployed workflows that quietly degrade because nobody is watching, measuring, or feeding corrections back.
Workflows that weren't viable six months ago are ready now — and nobody knows. The frontier moves faster than anyone can respond.
The pipeline does the deep work. The streams run alongside it — not after. At the centre, a living dashboard managed by agents.
Go deep only where the dashboard says it's earned. Every workflow that passes through makes the next one faster.
Assess the workflow as it actually runs — not how it's documented, how it's done. Fix it before you formalise it. The output is a machine-readable specification with defined success criteria, governance requirements, and the messy human context that makes it real.
Greenlight what to build next. The prioritisation matrix scores each workflow in a structured, machine-readable way — designed so agents can access, compare, and resurface candidates automatically. Then the collaboration model is set: which steps are AI-run, which stay human-led. Agents recommend. Humans greenlight.
Engineer the solution — build it, prove it, ship it. Every workflow passes through a parallel-run phase where the agent shadows the human before going live. Trust is engineered through evidence, not promises.
Nurture what's live. Monitor workflows in production — every human override is a data point, every correction is an instruction. The system learns and the specification evolves. This is how you nurture accuracy over time.
Track the frontier. New capabilities emerge constantly — Track watches for them and resurfaces workflows that weren't ready before. It also owns the decommission path. Without Track, the pipeline runs once and stops. With it, it compounds.
No large tooling migration. No advanced AI maturity required. Start with a recorded conversation, end with a compounding pipeline.
Agents handle the orchestration at each handoff. That's what makes this executable by a small team.
The framework is deliberately not a one-time engagement. It's designed to compound.
This isn't a finished product behind a paywall. It's a working framework — openly published, actively developing, and being tested across real organisations right now.
It's built from first principles, drawing on experience across digital transformation, product management, design thinking, and hands-on workflow mapping. It's currently being applied at two very different organisations, and they're hitting the same wall: not "should we use AI?" but "how do we actually do this properly?"
The framework will change. Resolution will sharpen as it meets more reality. Some parts are battle-tested, others are well-reasoned but unproven. That's by design — a framework that claims to be finished is a framework that stopped listening.
You're welcome to use it, adapt it, challenge it, and feed back what you find. That's how it gets better.
Every pipeline stage, both streams, templates, worked examples, and the thinking behind it. Openly shared and continuously updated.
Read the full framework