‘The accountability is with Mike’: Atlassian CEO’s AI agent rule

Atlassian cofounder Mike Cannon-Brookes wants one rule to survive the AI rush: the human stays on the hook. Speaking to Fortune at the Singapore Grand Prix, he said AI output should be treated as “Mike using ChatGPT,” not “ChatGPT doing it.” The point, in his view, is simple. Accountability has to sit with the end user.

That thinking shows up in Atlassian’s new “Agentic Multiplayer Protocol,” rolled out on Oct. 6. The company describes it as a control plane across its products, meant to turn AI agents into visible, governed members of a work team with their own profiles and owner attribution. In practice, an agent’s changes should appear in a document’s history the same way a colleague’s edits would. Cannon-Brookes said users need to know who did what and when.

He’s not impressed by the newer wave of personal AI agents, including Meta’s Muse and OpenAI’s Dots. Those tools, he argued, make the agent’s activity less visible at the very moment workplaces need more clarity around permissions, security, and compliance. A consumer app can be looser. A company system can’t be. If someone uses privileged internal documents to create something shareable, the leak risk is obvious and immediate.

Cannon-Brookes also thinks the bigger problem is that most AI interfaces are too clumsy for ordinary work. People end up using only the easiest features because the experience is poor. He compared that to stripping Microsoft Word down to keyboard shortcuts: the documents would all start looking the same. Atlassian has been pushing the opposite direction. In 2024 it launched Rovo, an in-house assistant meant to connect people, projects, code, and company knowledge across Atlassian and outside tools, with links to products like Figma, Codex, and Claude.

The company is also trying to prove that an Australian tech firm can play globally without pretending geography doesn’t matter. Atlassian, best known for Jira and Confluence, reported $1.8 billion in revenue in the quarter ended June 30, above the $1.7 billion analysts had expected. It also cut 10% of its workforce in March, about 1,600 people, as part of a shift toward AI and enterprise sales. Cannon-Brookes says Australia should focus on turning imported tech into economic value, not trying to build everything from scratch.