Reading Room

This is the exact accountability problem he will run into as he puts more of his own work through AI agents (AI systems that take several steps on their own): when one of them does something wrong, who actually owns that action?

Atlassian cofounder Mike Cannon-Brookes gave his answer to Fortune on the sidelines of the Singapore Grand Prix. He does not want "the AI did it" to become an office excuse. "That's ChatGPT on behalf of Mike. Like, that's Mike using ChatGPT. It's not ChatGPT doing it," he said. "The accountability is with Mike. It has to be with the end user, ultimately."

On October 6, Atlassian turned that view into a product. It rolled out something it calls the Agentic Multiplayer Protocol, a control plane (a system that sits above other tools and governs how they behave) that reaches across Atlassian's different platforms. It turns AI agents into visible, tracked members of a work team, each with its own profile and a named human owner attached. In practice, when an agent edits a document, that edit should show up in the document's history exactly the way a colleague's edit would. "You have to be able to see who did what when," Cannon-Brookes said.

He gave a small example of the failure he is trying to prevent: a system might log that an AI coding tool made a change, when in fact a person told it to make that change and should be named instead. "No, it didn't, someone did that," he said, insisting the record should point to the person, not the tool.

The idea is simple and still rare: treat an AI agent's work like a named employee's work, not an anonymous process. Most AI tools today log what happened, but not reliably who told it to happen. As more decisions and edits get handed to agents, that gap is exactly where "I didn't do that, the AI did" turns from a joke into a real excuse, whether inside a large company or inside a one person operation running several agents at once.

For someone directing multiple agents through different tools day to day, the practical test is the same one Cannon-Brookes is building into Atlassian's software: can you point to any change a week later and say which decision was yours and which step the agent took on its own? A name tag on every agent action is a small technical idea, but it is the difference between staying accountable for AI-driven work and quietly losing track of who decided what.

Outside your usual reading: a Reddit poster points out that among the world's top forty largest public AI companies by market value, not one is European, despite what the poster calls a deep pool of AI talent on the continent.

The gap between where the talent is and where the big companies actually got built is the whole story here: having capable people is not the same thing as building the most valuable company around them. That lesson holds well outside AI too.