Lab
Where sanctioned experimentation happens, and experiments are captured as portable bundles rather than left as one-off conversations.
AI Strategy · Framework 03
AI operating model
Once you know where the value is, how do you actually run AI day to day so good ideas become governed, durable capability instead of a drawer full of stalled pilots? This is the machine that does the compounding.
The thesis
Every AI workflow needs to find its correct operating layer. Some should stay personal. Some become governed team workflows. Some become formal business agents. Some become engineered production systems. For each one, decide which layer has enough governance for its risk and enough capability for its value, and move it there deliberately.
Two disciplines make it work. The unit you promote is a portable bundle, the prompts, skills, tools, eval set and governance captured together, not a clever chat. And the evaluation set, real cases with known-good answers, is the gate that earns production trust. Skip it and you are trusting a vibe.
The shape
Lab, Registry, Factory
One Lab, one Registry, one Factory, one owner. That is the operating loop, whichever products fill each box.
Registry
The version-controlled, reviewed library of approved bundles, with an owner and a design authority, so good work gets reused instead of reinvented.
Factory
The governed production environment where promoted bundles run under enterprise controls: identity, permissions, monitoring, audit.
The promotion ladder
Six stages from experiment to operated asset
A workflow climbs a ladder, and each rung has an owner and an exit gate. The gate criteria are what we build with you in an engagement; the shape is public.
Explore
An individual proves something is possible.
Validate
Prove the value and build the evaluation set.
Team Pilot
A defined team runs it under human review. The rung most firms are missing.
Harden
Compliance, data classification, access, exception handling, monitoring.
Promote
Move the bundle to the right production layer.
Operate
Monitor, measure and improve. There is no exit; this is the ongoing job.
The diagnostic
Five questions on how ready you are
They tell you how ready your firm is to run AI as a capability, not just try it.
- Capture. When someone builds something good, is it captured as a portable bundle, or lost when the conversation ends?
- Evidence. Do we have eval sets that let a non-expert trust a workflow, or are we trusting impressions?
- Reuse. Is there a registry of approved work, or does every team rebuild the same thing?
- Layering. Do we decide which layer a workflow belongs in, or promote everything or nothing?
- Ownership. Who owns the operating loop, and does governance travel with the work or get bolted on at the end?
Get the full framework
The full operating model: the portable bundle, the eval-set discipline, Lab, Registry, Factory, the six-stage ladder with owners and exit gates, the which-layer-is-enough decision, and the interview prompt that turns it into a tailored plan for your firm.
- The full framework write-up
- The tailoring interview prompt
Free. Goes straight to David. We only use your email to send the link and reply, no list and no spam.