DM Development

One human. Five tiny agents. One assembly line.

A tiny software factory that tests its own work.

Hand in a ticket and watch it go down the line. Archie writes the brief, Dev builds, Rex reviews, and Tess runs the tests for real, in a sandbox in your browser. Work that fails goes back. Work that is too big goes to the human.

Hand in a ticket

Brains

The board

Every station pins its note here as the ticket moves down the line.

    What just happened?

    The tests are real

    The brief ends with a definition of done: three checks that can be run. The page runs the code against them in a sandbox in your browser, with no access to this page, no network, and one second per test. A pass means the code passed, not that a model said so.

    Work can come back

    A factory needs inspection. Rex reviews against the brief and Tess tests against it. Work that does not pass goes back to Dev, and after a rework the tests run again. What still fails is not hidden: it is reported as a known issue.

    Small model, small jobs

    A small model cannot build your product. A station is at most one model call, with one narrow job and a strict format. Deciding who does what, what each step gets to see, and when a plain test beats a model's opinion: that is the actual engineering. Same job with bigger models and real systems, minus the trainee badge.

    The human behind the counter

    Hi, I'm Dominique. DM Development is my one-person software consultancy in Antwerp. There is no sales team and no account manager. When you work with DM Development, you work with me.

    Most of my time goes into client work together with AE, where I help teams design and build software that has to keep working long after the kickoff slides are forgotten.

    Lately a lot of that work involves language models: figuring out where an agent genuinely helps, wiring it into existing systems, and keeping it honest once it is there.

    What I actually do

    AI agents that do real work

    Designing and building LLM-powered agents and workflows: tool use, orchestration, evaluation and guardrails. Including picking the right-sized model, which is sometimes a very small one.

    Architecture that survives reality

    Many years of designing systems that have to scale, stay fast and stay secure, identity and access management included. None of that stops mattering when there is a language model in the middle: AI architecture is still architecture.

    The systems that build the products

    AI across the software development lifecycle: agents that write the brief, build, review and test, with real checks where a model's opinion is not enough. Much of my work goes here lately. The tiny factory above is the pocket-sized version.

    Need a slightly bigger factory?

    The agents are fun. The work behind them is what I do for a living. Tell me what you are building.

    info@dmdevelopment.com