This week saw major steps toward AI agent collaboration across different companies and tools. Microsoft and Google joined forces by adopting the A2A protocol, letting AI agents built on Azure and Copilot talk to other systems. IBM launched watsonx Orchestrate, a tool that helps businesses create AI agents in minutes using prebuilt parts. In the U.S., a new security-focused agent called TufinMate debuted, letting engineers manage networks through Slack or Teams chats.

SAS unveiled customizable AI agents that mix human and AI decision-making, with built-in rules to keep things safe. Experts highlighted that specialized agents (like ones made for specific jobs) work better than general tools, and teams of agents are becoming common. Developers are learning to build these systems at events like Microsoft’s Agents 101 Hackathon in San Francisco.