Skip to content
Smedja

How I work

First whether it fits. Then four steps. Then what it costs, who owns what and where your data goes. Everything below is what I also put in the contract, in plain words.

Is it a fit?

Usually yes for

  • Companies of around 20 to 250 people.
  • One process that demonstrably costs too much time.
  • Data that already sits in a system somewhere: a CRM, a mailbox, a web shop, an Excel sheet that is really a database.

Usually not for

  • A standard package that already does it. Then I say which one.
  • A whole ERP that needs replacing.
  • Work that needs a team of ten developers.
  • A report without a build. I write proposals, not advisory reports.
  • Data that does not exist anywhere yet. Then you start by collecting, not by building.

Is it about hiring or credit decisions? Stricter rules apply there. I raise that on the first call, not after it.

Four steps

  1. 1

    First call

    30 minutes

    You show me one process that takes too much time today. I ask questions, you prepare nothing. Afterwards you get an honest answer on whether custom software makes sense.

    From youHalf an hour and one example.

  2. 2

    Proposal

    Within a few working days

    What I build, what the build costs and what it costs per month to run. On paper, before anything happens. Nothing is invoiced before you approve.

    From youAccess to the system, or an export.

  3. 3

    First version

    1 to 2 weeks for most projects

    A working version on a copy of your own data. Your team tries it, I adjust it. Bigger or more complex? Then we agree a plan of its own up front, on paper too.

    From youOne fixed contact person and about an hour a week to test.

  4. 4

    In use and onwards

    After that

    Live, with a short guide and a handover. I keep it running, and we build on if you want to.

    From youTell me what could be better.

Why fast is not sloppy

AI saves me time writing code. That time goes into testing and into properly understanding your process. My own software runs on automated tests that run again on every change: Race Control on over 1,300, RoughCut and CreatorScope on over 700 each. Your project gets the same.

What it costs, in three separate lines

I only name an amount once I know what has to be built. How it is structured, I can tell you now.

What sets the amount

  • How many systems need connecting.
  • How clean your data is today.
  • How many people use it, and with which rights.
  • Whether it has to run on your own servers.

One awkward integration can cost more than a big but simple project.

  1. One-off

    Building

    A fixed price per phase, on paper up front. Nothing is invoiced before you approve.

  2. Per month

    Running

    Hosting, AI usage, monitoring and small repairs. The amount grows with the number of users and with how much goes through the system, because databases and AI calls cost more as more happens.

  3. Per month or per job

    Extending

    New wishes once it is running. Kept apart from running, so you decide how much, and so running never quietly gets more expensive.

When you get a number

After the first call, on paper: a fixed price for the build and an estimate per month for running. Term and notice period are in it, in black and white.

Who owns it

  • Code and accounts are in your company's name from day one. The source code, the hosting, the database, the domains. I work in them as a guest, not as the owner.

  • The rights are in the contract. Without an agreement, copyright in the Netherlands stays with the maker. So the transfer is in the contract in black and white, not in an email.

  • At delivery you get documentation and a handover. What it does, how it runs, where the keys are, and an hour in which I explain it to your people or your next developer.

  • If I drop out. Everything is in your name, so nothing stops. Any other developer can take over without having to call me.

Where your data goes

These are the providers I build with. Which of them take part in your project, and where their servers are, I put on paper per project before any data flows. I do not promise a region here; I agree it per project and check it then.

Hosting and codeVercel
The app itself runs at Vercel. The code lives in a repository in your company's name.
DatabaseSupabase
Your data sits in a database at Supabase, in a project in your company's name.
EmailResend
Emails the app sends (a notification, a draft, a confirmation) go through Resend.
AI modelAnthropic
If there is AI in it, the pieces of text the model has to read go to Anthropic through the API. Under its commercial terms Anthropic does not train on what comes in through the API (version of 17 June 2025, read by me on 30 September 2026). I re-read that for every project.
Your own email and calendarGoogle or Microsoft
If the app reads or writes in your Gmail, Outlook or calendar, it does so through a connection you allow yourselves and can revoke at any time.
On your own servers
Also possible. Then it counts in the price and in the plan.
Data processing agreement
If I process personal data for you, a data processing agreement comes with it. We sign it before any data flows.

AI, without surprises

  • Tools I build say so when you are dealing with AI. A draft email is called a draft, a summary is called a summary.
  • The tool does not make decisions about people on its own. Someone always looks before anything goes out.
  • At delivery your team gets a short guide: what the tool does, what it does not do, and what not to paste into it.
  • If it is about hiring or credit decisions, stricter rules of the European AI Act apply. I raise that on the first call.

How I use AI myself on this site and in my work is on the AI transparency page.

Show me one process that costs you time.

Half an hour, online or at your office. Nothing to prepare, and you are not tied to anything.

You will hear from me within one working day.