Who does the doing — and who keeps the judgment?
88% of organisations now use AI formally; 5% report transformative returns. The difference is the work mix: how much hands-on work AI actually executes, and how real the human loop around it is. Answer twelve questions and get your AI Work Mix score, the shape behind it, and the two moves that matter first. No email, no wait: the score is the mirror, not the sales pitch.
01 · The doing
How much hands-on work AI actually executes, across how much of the team, in what form.
01When a big chunk of hands-on work lands — drafting, summarising, data wrangling, first-pass code — how much of the typing does AI do today?
02Where in the team is that true?
03What form does the AI help take?
02 · The loop
What stands between an AI output and the outside world: who reviews, how deeply, and whether fixes stick.
04What happens before AI-assisted work goes to a client, customer, or colleague?
05When that check happens, what is it actually checking?
06When AI output is wrong and someone catches it, what changes?
03 · Decision rights
Whether it is written down what AI may decide or do alone — and where it must stop and hand to a human.
07In your team, is it written down what AI may do without a human?
08When an AI workflow hits something it is not sure about, what happens?
04 · Tool sprawl
Whether the AI tool landscape is a curated stack with defaults and owners, or an accumulation.
09How would you describe the AI tools your team uses?
10Two people on your team get the same kind of task — how many different AI tools do they reach for?
05 · Skills investment
Whether the team is deliberately getting better at working with AI — including the judgment half.
11What has the team invested in learning to work with AI?
12Are people getting better at the judgment half — spotting AI errors, knowing what to delegate — not just the prompting?
06 · Your work mix
Answer the twelve questions above — or load a typical team — and your score, the shape of your mix, and the two moves that matter first appear here. Nothing is sent anywhere unless you later ask us to email the breakdown; the whole rubric runs in this page.
How the scoring works
The third way is AI doing the hands-on work while humans keep the judgment. Your work mix is how much of the doing has actually been delegated, how intact the loop around it is, who holds decision rights, how coherent the tool landscape underneath is, and whether the skills to run it that way are being built on purpose. Twelve questions size those five dimensions; each answer is worth 0–100 maturity points and the score is the weighted average. Both halves matter: a team that delegates everything and reviews nothing is not further along than a team that delegates little and reviews well — it is differently exposed. That is why the result also names the shape of your mix, not just the number.
| Dimension | What it sizes | Questions | Weight |
|---|---|---|---|
| The doing | How much hands-on work AI actually executes, across how much of the team, in what form. |
| 3/12 |
| The loop | What stands between an AI output and the outside world: who reviews, how deeply, and whether fixes stick. |
| 3/12 |
| Decision rights | Whether it is written down what AI may decide or do alone — and where it must stop and hand to a human. |
| 2/12 |
| Tool sprawl | Whether the AI tool landscape is a curated stack with defaults and owners, or an accumulation. |
| 2/12 |
| Skills investment | Whether the team is deliberately getting better at working with AI — including the judgment half. |
| 2/12 |
The doing and the loop carry the heaviest weights, on purpose: they are the two halves of the third way itself, and the market's 88%-use / 5%-transformative gap lives in the space between them. Within a dimension, the question that moves the band most is the one that names a mechanism — a named reviewer who can block, a written no-go class, one default per task type — because mechanisms survive busy weeks and good intentions do not. The two recommendations are simply your two lowest dimensions, weakest first, with the answer that cost you the points.
What a score means — and what it does not
A high score is not more AI. Plenty of teams would raise their score by delegating less and reviewing better; the instrument is deliberately neutral about which half is weak. The four bands are descriptions of a system, not grades of a team: unmixed, emerging, deliberate, third-way rhythm — and the shape line matters as much, because over-delegated and under-delegated teams can share a number and need opposite next moves.
Three honest caveats. The weights are ours — reasonable people would nudge every row of that table, and the point is the conversation it starts, not the second decimal. The answers are yours, and self-assessment drifts optimistic on exactly the dimensions that are weakest: everyone believes their review is deeper than it is. And twelve questions cannot see your ticket queue, your repo, or what your juniors do at 5pm — a scorecard that pretended otherwise would be the least honest thing on this site.
We do not ask for an email to show you the score, because a number held hostage is a number we do not trust either. This is the same instrument logic as our shadow-AI exposure check and skills-gap checker: an honest, instant, ungated read — then the choice of what to do about it.
From a score to a working model
The first-move notes on your two weakest dimensions are things you can start this quarter for free. What they cannot do is install the habits at team scale — that is engagement work. Our AI upskilling engagement builds the judgment half role by role — review discipline, decision rights, error-spotting as a practiced skill — while workflow automation moves the doing half over properly, with human checkpoints where judgment belongs. Bring your scorecard to the first call and the framing conversation is already half done.
Or start with a conversation — the check is free either way, and nothing you answered leaves your browser unless you ask us to email the breakdown.
One email a month on practical AI adoption
What we ship for clients, what the tools can honestly do this month, and one thing worth trying in your own team — written by us, no hype, unsubscribe anytime.