Research · 17 October 2026
Choose your platform. Design your team. Put multi-agent AI to work.
With multi-agent AI, you can bring different capabilities together around the work you want to accomplish. A researcher explores a question, an analyst works through the findings, a specialist contributes knowledge, and a colleague helps shape the recommendation. Our new paper shows how to turn that possibility into a working team.
Making Multi-Agent AI Work is a practical guide for individuals, businesses, enterprises and AI leads who want to put teams of people and agents to useful work.
The opportunity extends across research, writing, analysis, development and continuing business processes. You can explore several questions in parallel, bring specialist perspectives into a decision, and connect contributions across different models and existing systems. People remain active members of the team, providing direction, expertise and judgement throughout the work.
A route from an idea to a working team
The paper helps you work through the decisions that make this possible:
- Explore the technology. Understand the different types of products, investigate those that fit your needs, and assess what a complete solution would require.
- Design the team. Define useful roles, decide how work and information move between participants, and make human involvement part of the design.
- Put it into practice. Build a small first implementation, learn from using it, and prepare it for regular operation.
A seven-step implementation cycle connects these decisions, from defining the work to measuring and improving the result. You can read the paper end to end or use it as a reference library as your implementation develops.
Guidance you can use
The paper brings together research and lessons from months of building and using our own multi-agent harness. Alongside the guidance are 111 requirements with best practice, software capability profiles, diagrams and 14 reusable working templates covering briefs, roles, handovers, evaluation and more.
It also connects to our Agentic AI Control Catalogue and Enterprise AI at Scale, giving you supporting detail for controls and wider organisational rollout.
Yours to use
The paper is published under CC BY 4.0. Read it, share it, adapt the templates and build on the guidance in your own work, with attribution.
Start with a piece of work you would like to improve — and explore what the right team of people and agents could make possible.
Multi-Agent AI · AI Agents · AI Operating Model · Agentic AI
Making Multi-Agent AI Work · edition v4.1 · 76 pages · published 17 October 2026 · sha256 8e61a5d3d2eded0944c7caa9cb19b67ebaf8f2054ac453ef8607e74d5542d1a2