Approach

Judgement over tools

Tools change every year; the hard calls don’t. Using the latest model is the easy part. The value and the risk both lie in the decisions around it: what to build, what to hand AI, what to keep human, and what happens when it’s wrong.

That’s the work we do: senior judgement about systems and AI, brought to the decisions that are expensive to get wrong. It’s the part a tutorial can’t teach and a tool can’t replace.

Judgement is experience, examined

Experience alone doesn’t improve it; experience examined against its outcomes does. That closed loop is what we bring, and what we build into your team.

The method

Five steps, used the same way on a stalled system, a standing advisory relationship, or an exercise in a workshop.

Frame→Options→Size→Decide→Record

Frame. What is actually being decided, and why now.

Options. The candidates worth taking seriously, including doing nothing.

Size. Size the consequences – three questions, every time: can we verify it, can we undo it, and what does being wrong cost?

Decide. Commit, and name the trade-off you accepted.

Record. Write it down, so the reasoning outlives the meeting.

Size is where most of the work sits, and where projects most often go wrong.

Where AI – or any new technology – belongs, and where it doesn’t

Every technology arrives claiming it changes the rules – microservices, serverless, the last framework, now AI. The sizing questions do not change with it: a technology is worth adopting only if it improves the system, not because it is new.

AI is the current hard case, because it makes delegating judgement cheap and pervasive.

The dangerous failure mode isn’t obvious garbage. It’s fluent, confident, subtly-wrong output that sails through a casual review.

Division of labor by stakes and reversibility A two-by-two matrix. The vertical axis is stakes, low to high; the horizontal axis is reversibility, hard to undo to easy to undo. High-stakes and hard-to-undo work stays with a human; low-stakes and easily reversible work is handed to AI; the two mixed quadrants are shared. Human decides; AI assists Architecture · security · what you can't reverse AI drafts; human verifies Consequential, but recoverable AI acts; human spot-checks Routine, but not easily undone AI has free run Boilerplate · drafts · breadth · first passes STAKES High Low REVERSIBILITY Hard to undo Easy to undo
The higher the stakes and the harder a call is to undo, the more a human stays in the seat.

Stakes and reversibility place a call on this map. Verification decides how far it can move: nothing gets a free run if nobody can check whether it was right.

Three lines we hold, and help you hold:

  • AI to make your people more capable, not more disposable. Adoption that hollows out your juniors or erodes your team’s understanding is a bad trade, however fast it looks on day one.
  • AI where it is worth its cost, not everywhere it fits. Some work shouldn’t be automated, and saying so is part of what you’re paying for.
  • Tools are examples; principles are the curriculum. Models and frameworks change; architecture, evaluation, boundaries and verification don’t. Pairing systems engineering with AI isn’t hedging – it’s the recognition that the architecture outlives the model.

The calls we help you make

The method is easiest to see in the questions it forces. Plenty of them have nothing to do with AI:

  • Is this a decision, or a preference with a diagram behind it?
  • What does this design commit us to for years, and what can we still reverse?

And some belong to AI alone:

  • Should this system answer, recommend, draft, or execute?
  • What has to be verified before output reaches a customer?
  • Where is human approval non-negotiable?
  • What should stay boring, deterministic, and deliberately non-AI?

We unpack how these get answered, piece by piece, in our writing. The fastest way to see it, though, is to bring us a decision you are still weighing.

How this reaches you

The same reasoning arrives two ways.

A live problem or decision

We work alongside your team – architecture, production AI, advisory – and make the reasoning explicit as we go, in decision records, evaluation criteria, and working practices you keep.

Consulting and advisory →

A capability your team needs

We teach the same reasoning through coached practice on your own projects and decisions, until it lives inside your people.

Workshops and briefings →

The difference is whether we make the call with you, or build the capability to make it without us.

Bring us a decision you're not sure about.

The fastest way to see how we work is to put a real, live call through it. Bring the messy version. We'll work it with you.

Start the conversation