Training

The agentic engineering kata

Duration
One day
Participants
5 to 15
Location
At your office
Language
Dutch or English

What the kata is

One day in a real codebase, learning step by step to direct an agent on code that already exists. Like a kata, you learn the movement by repeating it: every exercise builds on the code from the step before. The repository is the lesson, and at the end the build decides whether what came out is good enough.

How your day runs

The steps build on one another. Where an agent's answer comes from, through to what changes in your own role.

Start here

How the kata works, how to run the project on your own machine, and a first flag board: codes you earn with a solution that works.

How this kata works · The backend · Workshop

Context, model, mechanisms

Where an agent's answer comes from: what is in the context window, what each layer costs and what happens when it fills up.

Tokens · Your prompt · Tools · Context · The session · The harness · The model · Truth · Workshop · Recap

Agentic engineering

How you work with an agent rather than against one. The longest step of the day, and the afternoon workshop sits in it.

Project evolution · Project setup · Craft · Steering · Recurring patterns · Workflows · What it asks of you · Parallel workflows · Spending tokens · Workshop

Soft skills

The part that is not about code: what changes about the work itself, and about how you talk about it.

Change management · Expectation management · Impostor syndrome

What sticks

Every theme has an exercise behind it where you do it yourself, and every one of them comes back in the work the day after.

Truth You know how to check whether an answer is actually right, and where an agent stays quiet about what it guessed.
Workflow You choose per task how to hand work over: free prompting, a plan, a spec or an audit.
Cost You see what is in your context window, turn one interaction into money and know where the tokens go.
Setup You set the repository up with instructions, skills, hooks and scripts that come along in every session, so you do not type the same correction five times.
Parallel You put several agents to work at once, each in its own worktree, and find that your attention runs out before your tokens do.
Team You name what changes about your role, what you may and may not promise, and how you bring everyone else along.

How you practise

By doing. Most steps carry a flag board: short codes that only surface once you run the code on your own machine, instrument it or harden it. They are nowhere in the text and a search will not find them. The quizzes beside them run in the browser and keep working with the service down. In the afternoon workshop the build decides: coverage, complexity and mutation testing say whether your code is ready, not the agent's word for it.

What you bring

Experience and a laptop Programming experience, your own laptop with the IDE you know, and the rights to install tooling on it locally: the project runs on your own machine.
AI tools A working installation of Claude Code or GitHub Copilot CLI: an agent you run in a terminal. The kata starts at tokens and context, so experience with agents is welcome but not required.
Tech stack The exercises run on one stack, with the tooling that belongs to it. The principles and habits you take away fit any language.

Ready to make your developers work more efficiently?

Tell us briefly how big your team is and where it gets stuck today. We propose a day built around you, at your office, and we also say when a kata is not what you need.

Rather call? +32 11 11 10 20 Or mail [email protected] Beringen, Belgium

Fields marked with * are required.

Read our privacy notice to learn how we process your data.