From a single bot to collaborating agents: orchestration, tools and human oversight in one workflow.
Complex tasks require more than a single prompt: they require a division of labour. Agents with their own roles and tools, handing work over to one another — the way a team does.
The art is not more agents, but the right direction: who does what, when does a human step in, and how do you keep sight of quality and costs? That orchestration craftsmanship is what you learn here.
This subject did not exist five years ago, because the object of design is new. You are not designing software but a division of labour: which role goes to which agent, who hands off what, and when a human takes over again.
That is organisational design that happens to be written in code. So the hard questions are not technical ones: where do you place authority, how do you stop one error multiplying down the chain, and what may an agent decide alone with nobody watching.
You build a working multi-agent prototype for a real process from your own organisation.
A sales engineer has three agents work together: one gathers requirements, one calculates, one writes — a human approves.
A policy adviser builds a workflow that finds sources, summarises them and flags contradictions for human review.
A team coach has agents classify tickets and propose solutions; borderline cases are escalated automatically.
You take this module the way you take the whole programme: classes every other week on Friday and Saturday, with a study load of 15–20 hours per week, of which 10–15 hours are self-study. The teaching is a mix of classroom lessons, practice-based learning, blended learning and working groups or study teams — taught by lecturers who practise the profession themselves every day.
You conclude each theme with a professional product or a technical solution based on a real situation in your own work, which you discuss in an assessment with the lecturer. This way your portfolio grows with real work — and your employer benefits directly.
The format is not new: the teaching approach and the AI modules have been running for three years at the AI Proeflokaal — with business, with government, and across part-time modules and minors. What you meet in class here has already been worn in elsewhere.
After this module, a multi-agent prototype is running for a real process — with checkpoints and a cost overview.
Email or call the programme — we're happy to help you think it through.