Essay + instrument · July 2026 · 9 min
AI won’t take your job. It will take your queue.
A working, priced, animated model of an eighty-person IT department — and what actually happens to the machine when you let an AI into it, one role at a time.
Ten Days · AI transformation
You already know the answer. Their day rate clings to a pre-AI world. Their promise sells a future that contradicts the price. So either the technology they recommend does not work — or it works, and the saving is stopping at their margin.
The discovery: eight hours, from £300, on a project you already have.
The flow behind this page is live — it reacts as you scroll ↓
02 · The hours
An apology first: we are about to lead with costs, and it is crude. But the new value point of this work is largely unknown, and it needs putting in front of you plainly.
Two shapes of work, from our model of an ordinary department. A feature spends thirty weeks getting from asked-for to live, and two-thirds of that is waiting. A medium issue — identified, fixed, tested, released — is fourteen elapsed days and about £4,500 through people. Through an agent with a human review step: four hours and about £30, papers in order. The hours the work needs have not drifted down. They have collapsed.
Our trade is implementation — realising those hours inside your own business: finding where they hide in your queues, turning them into capacity, shortening the road from spend to value. Time-to-value is the number we move.
Every consultancy should be doing this. The method is not a secret.
One feature
30 weeks
from asked-for to live in an ordinary department — two-thirds of it is nobody doing anything
One medium issue
£4,500 · 14 days£4,500
identified, fixed, tested and released — through people, then through an agent, as our model prices it
The waiting
100 days100 days
days that feature spends waiting on handoffs, before and after — modelled, not promised
These numbers are live — open the model and argue with them →
03 · The way in
Discovery is eight hours of our expertise against your numbers, and you choose what it prices. A view of the effort to complete a project you already have in flight, or nearly have. Or a view of where AI fits your own processes, and what applying it would return.
We suggest starting with the project. It is concrete, its history is known, and the answer is useful the day it arrives. You keep the model and the figures either way — and at worst they are a price comparison that whoever bids for the work next has to beat.
04 · The conflict
A decades-old consultancy, tens of thousands strong, billed by the hour, cannot sell you fewer hours.
Ask your consultants what their own AI expertise has saved them.
05 · Our answer
We implement. Experts directing AI delivery teams — solutions built, infrastructure scripted, AI systems put to work — with a human answerable for every step. Our prices dropped because the hours did, and the saving is passed on: by our own arithmetic, a £4,500 fix becomes a £30 one, in the price you pay.
Start with the needs you already have — the system to finish, the integration to fix, the next version to build. AI enablement comes second, and comes easier: by then you know what a build costs before anyone quotes you.
A stalled project is the most straightforward case. We call that engagement a treatment — the ten days work pointed at the project’s arrears to restore good order: catching up to your own timeline, then we leave, and you’ve a better idea of AI abilities within your remaining technical or project procurement staff.
06 · Start
The whole of it: name the challenge, give us eight hours from £300, and hear what finishing it costs at agent rates — model, figures and every assumption included. If nothing is worth doing, you hear it straight. If something is, the ten days implement it.
Three lines is enough — where the work waits, what different would look like, who owns the decision. Replies come from principals, usually within a working day.
Proof, not promises
You should not take our word for any of this — a firm’s opinion of its own value is not evidence. So we publish our working. The instruments below are live, priced models you can run, re-parameterise and disagree with.
Being the person in the room who has argued with the parameters is rapidly becoming the whole job.
Essay + instrument · July 2026 · 9 min
A working, priced, animated model of an eighty-person IT department — and what actually happens to the machine when you let an AI into it, one role at a time.
Instrument · July 2026 · Live model
A firm never chooses make or buy — it runs a mix. Watch what the mix does when AI drops the cost of knowledge work toward the cost of compute.
In the workshop
Coase’s question with a dial on it: drop the cost of coordination and watch where the seam between firm and market moves.
The constraint migrates whether or not anyone in your building is watching. Eight hours, from £300, prices the challenge you already know by name — then you decide. Most people who read this far already know which project they would name. Name it.