Agent Build · 2 weeks · €3,000 fixed
Your first agent, working in two weeks
One task, chosen in the first days with the payback math in the open, then built and running on your own data — an agent doing the work, with a number on how often it gets it right and a list of what it does not.
Getting an agent to work once is the easy half
The expensive failures are not model choice. They are the wrong task, no agreed definition of a right answer, and a demo that was never going to survive contact with real inputs.
The demo that never became a system
It works on the rows someone picked for the demo. Then it meets the inputs nobody thought to try, and the difference between the two is where the project actually lives.
Nobody agreed what a right answer is
Without a set of real cases and the answers written down in advance, “it works” is an opinion, and so is “it got worse after the last change”.
The wrong task went first
Something that runs once a month for an hour is a bad first agent no matter how much it annoys everyone. Picking is arithmetic, and it takes days rather than weeks.
An agent you can run, and a number you can check it against
Two weeks from kickoff to the walkthrough, with a go/no-go on day three and about three hours of your team's time along the way.
- A working agent
- One task, running on your own data at the end of the fortnight, in the mode where the agent drafts and a person approves.
- The number on how often it is right
- Measured on a set of your real cases before handover, not on rows we chose.
- The list of what it gets wrong
- The cases it fails, the ones where it should hand back to a person, and the ones where we would not trust it at all. This is the deliverable that argues against buying more from us, which is why it is on the list.
- The payback math, written down
- Which task we picked and the assumptions behind it, in the open, so you can argue with the arithmetic rather than with our conclusion.
- All of it is yours
- Code, test set and measurements. Carry on with us, your own team, or anyone else.
What the fortnight costs and what happens in it
Fixed price, so there is no scope creep on your bill. For scale: published AI consultancy rate cards put a two-to-four-week proof of concept at £15,000–£40,000. This is less because it is less — one task rather than a programme, one senior pair rather than a team, and the data you already have rather than a pipeline built to feed it.
- Days 1–3: two one-hour interviews, sample data, and a look at the systems involved. We pick the task and write out the payback assumptions
- Day 3: the go/no-go. If nothing here pays for itself, we say so and there is nothing to pay — not for the three days, not for the interviews. Putting the exit on day three is what makes that affordable for us, and it is the only point where saying no is still cheap for both sides
- Days 4–6: the test set. Real cases from your work with the correct answers agreed up front, which is the step almost everyone skips
- Days 7–9: the agent, built against that set and run on your data
- Day 10: walkthrough. The agent, its score, and an honest list of what it still gets wrong
Fixed against one task and the data you already have. If the go/no-go on day three comes back no, there is nothing to pay: the fee buys a fortnight that produced a working agent, not the days it took to find out that it would not. If your case needs an integration written from scratch, we say so on the free intro call, before you have paid anything.
What is not in the fortnight
- Production hardening and high-availability deployment — the agent runs on your data, not yet under your uptime commitments
- Autonomy without a person approving the output. Draft-and-approve is the mode we hand over in, deliberately
- Integrations that have to be written from scratch against a system we have not seen. We say on the intro call whether your case needs one
- Roles, permissions and audit logs for multiple users
- Monitoring dashboards and alerting on the agent's own behaviour
- A second task. The fortnight is fixed against one
Anything on this list is a separate decision with its own fixed price, agreed before the work starts rather than invoiced afterwards. Most of it is also cheaper to judge once the first agent is running and there is a number to argue from.
Price
€3,000 fixed
2 weeks, kickoff to final walkthrough.
When not to buy this
Four cases where the honest answer is that you should spend the money elsewhere, or not spend it at all — we would rather say them here than on the call. Flip them and you have the buyer this is built for: one task that runs often enough to matter, the data it runs on already in your systems, and somebody on your side who can say what the right answer is for fifty real cases. If those three hold, two weeks is enough.
Your processes live in people's heads
Then the first job is writing one of them down, and that is a fortnight of your own operations lead's time rather than €3,000 of ours. It also makes everything after it cheaper. Come back when one process exists on paper.
You already know exactly what to build
If the spec is settled and what you want is hands, you would be paying for a fortnight of judgement you do not need. Hire a contractor and give them the spec.
What you want is a chat widget on your website
That is a solved, productised category with good off-the-shelf options at software prices. Buying it at consulting rates is bad for you and dull for us.
You need something to show a board that will not act on it
What comes out of this is a running agent and a number, not a document to present. If the decision has already been made upstairs, the number will not change it and you should not pay for it.
Agents we have already shipped
Each of these went the same way: one job, real data, and a way of telling whether it was doing the job right.
In production
Catalogue-to-Offer Agent for a Uniform Manufacturer
A production agent that generates only from real catalogue items, shows its plan before spending, and logs the provenance of every image it delivers.
Read the project →In production
Country Explorer: Location Intelligence for Restaurants
A team of agents reads location, footfall and demographic data and writes sourced expansion briefs a human can check against the figures behind them.
Read the project →In production
LetAI: Nutrition Estimation Agent in Production Chat
The agent only reached production chat after its evaluation pipeline was expanded, with new datasets and broader case coverage added before release.
Read the project →
Frequently asked questions
That is the right question to ask, and we would rather answer it than pretend the incentive is not there. Three things hold against it. The assumptions behind the payback math are written down rather than summarised, so you can check the arithmetic instead of trusting the conclusion. The go/no-go sits on day three, and if it comes back no we are not paid — not a reduced fee, nothing. That is the part of this answer that costs us rather than you, and it is the reason the payback math sits at the start of the fortnight instead of being sold as its own service. And the number we hand over at the walkthrough comes from your cases, not ours, so it can embarrass us. If you would still rather the recommendation came from someone with nothing to gain from it, that is a fair thing to want — buy the analysis elsewhere and bring us the conclusion. We will build against someone else's brief.
Fixed price on exploratory work usually means one of two things: the vendor padded the number, or they narrowed the deliverable until it was safe. We did the second, deliberately, and the narrowing is written down rather than hidden — one task, your existing data, draft-and-approve rather than autonomy. What we did not narrow is the measurement, and that is the part that can go against us. A narrow thing that is measured is worth more than a broad thing that is asserted.
Real enough to do the work on your data and be measured doing it. It is one task, and it runs where a person still approves the output. What is not in the fortnight: production hardening, wider autonomy, and integrations that have to be written from scratch against a system we have not seen. Those are a separate decision, and we say which of them your case needs before the start rather than at the walkthrough.
Then we say so in the first days, with the math, and you pay nothing at all — not a reduced fee for the days we used, nothing. The write-up is still yours: what would have to change for the answer to become yes, the data you would need to start keeping, the volume at which it starts to pay, the step that would have to be written down first. A no you can act on is worth having, and this one costs you the three hours of your team's time it took to reach.
About three hours of your team's time in the first days, a look at your data and systems, and — this is the part that decides the quality — someone who can say what the correct answer is for the cases in the test set. We do not need production access for the fortnight.
PhD-trained researchers and engineers, the same people who ship our projects. No account managers in between. You can see who they are on the team page.
Whatever the number says it should. Usually one of three: widening the agent's autonomy where the measurement supports it, putting your documents behind it so it answers from what the company knows, or wiring the test set into your releases so a drop in quality shows up before your users find it. Each is a separate decision and a separate contract. We will not put a price on any of them here, and the reason is not coyness — what widening autonomy costs depends on the number the first agent scored, and quoting it before that number exists is guessing. What we can say is the shape: these are single-task-sized pieces of work rather than programmes, and the second one is cheaper than the first because the test set already exists.
Want to see who you'd be working with? Meet the team.
No commitment to continue
Request a build
Tell us what the agent would have to do. The first 30-minute call is free: if the task wouldn't pay off, we'll say so before you spend the €3,000.