Analysis
Atlassian co-founder Mike Cannon-Brookes says the blame for an AI agent's mistake always belongs to the human who deployed it, not the model. "The accountability is with Mike. It has to be with the end user, ultimately," he told Fortune on the sidelines of the Singapore Grand Prix.
The comments follow Atlassian's October 6 launch of the Agentic Multiplayer Protocol, a control plane that gives AI agents visible profiles and a named human owner across Jira, Confluence and the rest of its platform, according to ITDaily. Rovo, the company's in-house AI assistant launched in 2024, already connects people, projects, code and company knowledge, letting users link accounts with Figma, Codex or Claude and switch between them.
Why consumer-style agents worry him
Cannon-Brookes drew a sharp line between consumer AI agents and workplace ones. He pointed to Meta's Muse and OpenAI's Dots as personal agents that hide what they're doing behind the scenes -- fine for a consumer app, he said, but unworkable at work, where an agent quietly building a chart from internal documents could leak information without anyone noticing. He also pushed back on a system that credited an edit to "Codex" rather than the person who made the change, arguing that agent work should be traceable like a colleague's, with a visible record of who did what and when.
The accountability push comes as Atlassian's own numbers get scrutinized. Reuters reported $1.8 billion in revenue for the quarter ended June 30, above the $1.7 billion analyst average, even after the company cut about 10% of its workforce -- roughly 1,600 people -- in March as part of a shift toward AI and enterprise sales. Rivals including Microsoft's Copilot, Salesforce's Agentforce and ServiceNow are racing to ship their own agent-identity and governance layers, making accountable-agent tooling a competitive feature rather than a compliance afterthought -- part of a broader pattern Pulse has tracked in Atlassian's other 2026 moves.
For founders selling AI tools into the enterprise, Cannon-Brookes is previewing the next procurement question: not whether an agent can do the work, but who is named as responsible when it's wrong. Startups building audit trails, agent identity or attribution layers now have a reference customer-side argument for why that feature matters, straight from a company running Jira and Confluence inside most large enterprises.
The limitation is that Atlassian's own framework still depends on administrators correctly scoping agent authority in the first place -- a governance tool is only as good as the permissions someone remembers to set. Cannon-Brookes also conceded that most people still only use AI tools for basic tasks because the interfaces are hard to navigate, comparing it to early word processors before toolbars made features discoverable.
Atlassian, founded in Sydney and listed on Nasdaq since 2015, now has employees in 49 of 50 U.S. states -- scale that makes its stance on agent accountability less a philosophical position than a product requirement for the enterprises it already serves.
