The AI Office

AI is becoming easier to access. The harder questions are changing.

Not a feed of tools. A working office: how I think with AI, how I build with it, and how I decide when not to.

AI is where the seeing gets demonstrated. It is not the point.

The stance

I don't start with AI. I start with what is actually happening.

Then, if AI helps us solve the problem faster, better or differently, we use it. If it does not, I say so, and we do the thing that works instead. That is the whole position, and it is a rarer one than it should be.

Most of the AI conversation I hear in boardrooms is a conversation about tools. Which model. Which vendor. Which pilot. The smartest leaders I speak to are not trying to become AI experts. They are asking a smaller and much harder question: what does this change about how we decide, what we own, and what we no longer need to be good at ourselves?

AI is an accelerant and a capability. It was never the point.

So the questions in this office are not about features. They are about leverage:

How I think with AI

Three frameworks I keep coming back to.

These are not products. They are the shapes I use to think, and I am sharing them because a good frame is worth more than a good tool.

Framework one

The AI Value Ladder

Most people are standing on the first rung and calling it the whole ladder.

Every rung is a different relationship with the technology, and a different kind of value. The higher you climb, the less it looks like using a tool and the more it looks like owning a capability.

  1. AskA question in, an answer out. Useful, and where almost everybody stops.
  2. RememberIt knows your context, your history, your standards. Now the answers are yours, not generic.
  3. ThinkIt reasons through a problem with you: options, trade-offs, what you have missed.
  4. CreateIt produces the artefact: the analysis, the draft, the model, the working prototype.
  5. ActIt does the thing, inside your systems, with your permission and your guardrails.
  6. CompoundEach cycle leaves something behind that makes the next one better. This is where the leverage lives.

Framework two

The Orchestration Loop

One model is an assistant. A directed set of models is a team.

Stop asking which model wins. The interesting move is to stop choosing one assistant and start designing an AI team, with distinct roles, a critic that is not the author, and a human who decides. The loop is the same whether it runs on Claude, on ChatGPT, or across both.

  1. FrameState the problem, the audience, the constraints and what good looks like. Most bad output is a bad frame.
  2. GenerateProduce a first answer, or several from different angles.
  3. CritiqueA separate pass, ideally a separate model, whose only job is to find what is wrong.
  4. ImproveRework against the critique. Repeat until the critic runs dry.
  5. DecideA human chooses, owns the decision and carries the accountability. Always.

Framework three

Rent the intelligence. Own the context.

If intelligence becomes rentable, what becomes valuable?

Foundation intelligence is now something you rent, by the token, from a handful of suppliers, and your competitor rents exactly the same. So the advantage cannot live there. It moves to what only you have: your context, your memory, your workflows, your rules, your judgement, your customers, and the way your people actually make decisions. Value moves to what only you know, and to how well you own it.

What you rentReasoning. Language. Pattern recognition. Speed. Breadth. Everything a large model does the moment you ask, from whichever provider is best for the job this year.

What you ownYour organisational memory. Your proprietary context and data. Your workflows, permissions and rules. Your institutional knowledge. Your customer understanding. The way you evaluate good from bad. The way your best people decide.

Privacy matters, and it is one dimension of this, not the whole of it. Ownership is the wider idea: knowing what is yours, keeping it in a form you control, governing who and what may act on it, and being able to plug it into any intelligence you choose to rent, without being captive to one.

The businesses that win this decade will not be the ones with the best model. They will be the ones whose own context is the best organised, the best governed and the best connected to the intelligence they rent.

Built, not described

I don't just talk about what AI might do. I build with it.

I work directly in the Claude and OpenAI APIs and have designed, shipped and run AI platforms of my own. So when I say something can be made to work, that is not a slide. It is something I can prove by building it.

The pattern is always the same: I see something, then I build something. The building is how I know the seeing was right.

See the work →

From The Unseen

Reading for leaders who would rather understand than adopt.

More in The Unseen →

The Business X-Ray

Wondering what this thinking finds in your business?

That is the Business X-Ray. Give me your website and I will look at your company from the outside: what it says it is, what may be happening underneath, where value looks underused, and where AI would help and where it would not. Every X-Ray is reviewed personally before it goes out.

Show me what I'm not seeing →