Agents have been the big wave for the last year or so, but the real power comes when agents can operate semi-autonomously – not prompted by a human at every step, but able to complete more tasks on their own and check in for approvals as needed. This is the real step change in AI – the last 30 years have seen us create apps to be used by humans, but now we can create autonomous agents that can automate complex processes via deep reasoning, connections to data, and the ability to use tools like a human does, including browsers, applications, web searches and more.

In Copilot Studio agents, Microsoft have invested heavily in autonomous capabilities in the last year. This session covers the fundamentals of what’s possible, then gives an update on the latest enhancements to the toolkit across agent instruction processing, knowledge integration, tool use, and multi-agent flows.

But can you really say “follow up on each marketing lead by sending a personalised e-mail to the client and update our back-end systems appropriately” and AI magically does all the right things? Where are the boundaries? How autonomous can autonomous agents be? In this session we’ll show updated top techniques for success, and where you need to draw the line between instructions and specific items for your agent to call.