Atlassian CEO says each AI agent must have a human owner

Atlassian cofounder Mike Cannon-Brookes told Fortune that a person, not the AI, is the owner of the result. On 6 October, Atlassian released the Agentic Multiplayer Protocol. It is control software that makes agents members of work teams that all can see. Each agent has a name and an owner. Atlassian gave no test results or user numbers.

How it works

The edits of an agent show in the document history, as the edits of a colleague show. Cannon-Brookes said that all must see who did the work, and when. Software must not show that Codex made an edit. A person made it.

The proof

Atlassian gave no test results, no customer numbers and no data on the protocol. Only the CEO gives information about it.

What is not known

Cannon-Brookes said that agents for one person, such as Meta Muse and OpenAI Dots, do not show their work. This can work for consumers, but a company must have rules on permissions, security and compliance. It is not known if other firms will use the protocol.

Insight: Many firms now write the rules for agent ownership

Atlassian released its protocol on 6 October. Meta and Sierra work on an open protocol for agents, with Shopify, Stripe and Walmart as partners. Jessica Blue of Money20/20 says no one can show who approved a payment agent. Each company seems to want its record of approval to become the standard. The first record that many companies use can set the rules. Leaders must select their record before a vendor does.