Never write alone, Rovo Agents

Rovo Agents brought customisable AI agents into the Editor across Jira and Confluence, announced at Atlassian's annual conference, Team24 and reaching general availability in September.

Guiding principles

Working in a new space, I took a step back to develop principles that would guide design decisions. A key starting point was teasing apart general AI from agents.

Personified AI The teammate mental model sets clear expectations, while remaining honest that this is artificial, not a person.

Collaborative Multi-turn chat is a powerful way to help users navigate ambiguous problems. Where the problem isn't ambiguous, a bot or automation is often the better answer.

Suggestions Agents have no agency to complete tasks unilaterally. Trust is everything, so users need to always be in control.

Vision

Never work in isolation — instead collaborate with responsive AI agents that provide immediate feedback, thoughtful guidance and support.

The vision was a multiplayer-style UX in the Editor, where agents felt like real collaborators rather than tools you invoke.

Design direction

Following the principles, the focus was on making collaboration with agents feel natural, the same way you'd work with a teammate.

That meant extending Edit Suggestions, typically used for human edits, to support agent changes. A human still accepts or rejects every suggestion. Agents also bring specific skills and knowledge that can be called on directly within the editor.

From there, agents were designed to provide additional context and support when teammates are unavailable, commenting, linking to relevant knowledge, and surfacing information from across the organisation.

To meet the GA milestone, the command palette was extended to become the Agent Palette. A chat integration allowed users to move content fluidly between the chat sidebar and the editor.

Outcome

The work went from zero to a testable demo within a few weeks, giving real users the chance to provide feedback before general release. Edit Suggestions became a new Editor feature designed specifically to support agent use cases. The mental model and interaction foundations established here are now being used to build the next iteration of the product.