A team of AI agents, each with one job, is easier to steer than one agent that does everything. At aiblLIVE London 2026, Oliver Given of Google ran a hands-on workshop building a travel-planning multi-agent system with Google’s open-source Agent Development Kit.
This guide turns the session into a design method a business leader can run on a whiteboard before anyone writes code, plus the build steps for the technical team.
As presented in the workshop by Oliver Given, Google
A small team of AI agents where each member has one job and a root agent decides who does what. Instead of one agent with a long list of instructions, specialists handle each type of request and the root agent hands work to the right one.
List the types of request, give each a specialist agent, add a root agent that delegates, decide what must be remembered between agents, pick a workflow pattern for multi-stage work, match the model to the job, test by conversation and fence each build in its own project.
Sequential when each stage needs the output of the one before. Loop when a critic agent should review and send work back to improve. Parallel when stages do not depend on each other and speed matters.
The full workshop guide gives you:
From a hands-on workshop at aiblLIVE London 2026, run by Oliver Given, AI GTM Lead at Google. See all the workshops and boardroom roundtables from the day.
