What I offer · AI that does the work

We want to do something with AI. But what, and where do we start?

AI is moving from assistance to execution: from a helper that makes a person faster to a system that takes over defined work, checks it and escalates when judgment is needed. I design and build those systems, connected to the software, data and people already in place. This page guides you to the step that fits your situation.

The demo was impressive. Then it went quiet.

AI needs a real role in our product, not another feature.

More people, more handovers, more manual checks. It is stuck.

01 · Three situations, three routes

Where are you?

Three situations in which people call me, in their own words. Recognise one, then follow that route: every route ends in a concrete offer with a price attached, and on every page you first read why things stall there and only then what I offer. Recognise none of them? Then there is a fourth door.

A red box on a wooden trestle, with a tag on a thread.

From pilot to production

The demo was impressive. Then it went quiet.

As soon as it comes down to reliability, existing systems, data and accountability, things stall. Usually the urgency is there and so is the budget. What is missing is someone willing to make the move to production.

See The Sprint

A blue appliance with an open compartment in its front, holding a small red robot as a built-in part.

AI inside your product

AI needs a real role, not another feature.

The question is rarely whether it can be done. The question is what role AI plays in your user's experience, what may run autonomously and where your team stays accountable.

See AI in your product

A tall stack of paper with one red folder sticking out.

A knowledge process is stuck

More people, more handovers, more manual checks.

Searching, reading, comparing, entering, preparing, passing on. The pain is concrete: capacity, lead time, quality, or dependence on scarce people. We redistribute the work instead of speeding up one step.

See a stuck knowledge process

A spotting scope on a wooden tripod, with a red band around the scope.

Not sure which situation?Then The Scan is the first step. Two working days with the people who know the work, to find out where judgment sits and which part has to be proven. Half credited if a Sprint or a Build follows. Book a call and we schedule it.

Book a call

02 · Seven kinds of work

What can you trust AI with?

AI execution stays abstract until you see which work it is about. This is the kind of task a system can take over, with a human on the decisions. If your work is not in there, you would rather hear that now than after a quote.

Where my work starts and stops.

What you can trust AI with

Read a request from email and attachments, check it and create it as a case file.
Answer a customer question using CRM, order system and knowledge base, and update the case file right away.
Prepare a quote from customer agreements, product data and rates.
Collect documents, compare them and request what is missing.
Investigate, enrich, prioritise and route tickets.
Move data between systems that do not talk to each other properly.
Give AI a real function inside an existing software product.

You do not need me for this

Is there already a good off-the-shelf solution? That is usually faster and cheaper than bespoke.
A chatbot on your website or in your app.
A general AI training that stands apart from your own systems and data.
An organisation-wide AI strategy or adoption programme.

Do you work in software, industry or professional services?

The For whom page shows, per market, which work people there still move by hand between systems, and when I am and am not the right fit.

See For whom

03 · Six things, three steps

What can you buy from me?

Six things, in three steps: first decide what has to happen, then build, then stay involved for as long as you want. The Scan when the question is not sharp yet, The Sprint as the usual way in, and if you already know what you want you can go straight to The Connector.

First

Deciding

Working out what has to happen, and whether AI can carry this work.

The Scan

two working days · fixed price

€3.500

A pair of binoculars on a wooden tripod, with a red band around the barrel.

For when you know something has to happen but not yet what. Two days with the people who know the work, one first trial on the part where the doubt sits, and a priced proposal for the next step. Half credited if a Sprint or Build follows.

When it is not sharp yet See The Scan

The Sprint

AI Execution Sprint · ten working days · fixed price

€12.500

A small device under a glass bell jar, with a red tag on a thread.

From a prototype gathering dust to a production decision you can defend. One bounded question, working code on the critical part, and a build-or-no-build recommendation with a production path.

The usual way in See The Sprint

Then

Building

Building the system, or the part that is still missing in between.

The Connector

an MCP server on your own system

design €2.500 · build from €12.500

A pipe coupling with a red valve wheel on top.

Your own software usable in any AI assistant, with fixed gates and the permissions your people already have. Runs on your environment, without rebuilding your application.

Fast lane See The Connector

The Build

from €20.000

A scaffold of blue tubes with one red plank on the top level.

If the direction is right, I build the system or lead the implementation together with your team. Not a standalone demo, but software for the real environment: data, integrations, business rules, approvals, monitoring and exceptions. A Build starts with a fixed scope and is then quoted in phases.

After the Sprint See The Build

After that

Staying involved

Someone who stays involved once it is running.

Fractional AI Architect

two to four days per month

€2.500 / €4.500 per month

A drafting table with a rolled drawing and a red T-square.

Structurally involved as a senior partner: assessing opportunities, guarding architecture, building critical parts, coaching teams. Usually what follows a Sprint or a Build, not something you start with.

After a Sprint or Build

Maintenance and safety net

cancellable monthly

from €350 per month

A toolbox with the lid ajar and a red spanner inside.

Only worthwhile once something is running. Updates when the standard changes, a fixed number of hours per month for extensions, and access to parts I already built for others. Cancel, and everything simply keeps running.

Once it is running See The Build

04 · Start small, prove it quickly

How does it work?

The same three steps in every route: scope first, then prove, only then build. From the smallest step to a full Build, without slides and without a demo that falls apart in production.

Step 01

Scope it

A drafting compass with a red point on a folded drawing.

One question, sharply bounded. What has to be executed, what is the critical part, and where does judgment stay with a person?

Step 02

Prove it

A small gearbox on a test block, with a lit red lamp.

Working code on that critical part, in your environment, instead of slides or a demo that falls apart in production.

Step 03

Build and hand over

A Y-shaped pipe with one branch shut off by a red valve.

From proof to system, embedded in your existing landscape. With monitoring, guardrails and a team that can run it themselves.

05 · What is running

What can you verify?

AI execution is a new part of what I offer, and I do not have a paid client case in it yet. The technology and its parts do already run in production, and everything below is public or can be seen live.

The Connector runs in production at Careerguide.

An MCP server on Careerguide's system, with fixed gates and the permissions the people there already have. I show it live on the real system.

See The Connector

Shown live in a demo

A conveyor belt with three boxes, the red one just rolling off.

I run my own delivery pipeline.

Agents pick up tasks from my backlog, open their own branch and deliver the work as a pull request. I specify, review and merge. The quality gates are not something an agent can skip: test suite, code style, front-end build and a separate review pass.

How the delivery pipeline works

Described in the Lab

A ruled sheet of paper with a round red stamp in the corner.

MCP E-Invoice EU is public.

An MCP server that makes the European e-invoicing rules (Peppol, EN16931) usable by AI systems. Open source, so you can read the code, the decisions and the edge cases yourself instead of taking my word for it.

MCP E-Invoice EU on GitHub

Open source on GitHub

A round gauge on a stand, with one red needle.

Cranely proves the other side.

A scorecard tool where judgment deliberately stays with the human. Not everything should be delegated, and working that out is the work described above.

View Cranely

Try it yourself

Sitting on an idea, a prototype or a stranded experiment?

In half an hour we'll know whether this is work AI can be trusted with. And if it isn't, you'll hear that too.