Atlassian Rovo: Putting AI Agents to Work Across Jira and Confluence

Rovo layers AI search and agents across the whole Atlassian graph

Atlassian Rovo is Atlassian's AI layer that sits on top of your organisation's knowledge — Jira issues, Confluence pages, JSM tickets, and connected third-party tools. It has three pillars: Search that understands natural language across everything, Chat that answers questions with cited sources, and Agents that actually do work. For a platform engineer, the interesting part is the agents. This is how I think about deploying them.

The Rovo Knowledge Graph

Rovo's value comes from its teamwork graph — a connected map of your issues, pages, people, and the relationships between them. Because it indexes across products and even external tools like Google Drive or GitHub through connectors, a Rovo answer can pull from a Confluence runbook, a Jira incident, and a Slack thread at once, with links back to each source. This cross-product context is what a plain chatbot cannot do.

Rovo Chat: Answers With Receipts

Rovo Chat is a conversational assistant grounded in your data. The critical difference from a generic LLM is citations — every claim links to the page or issue it came from, so an agent can verify before trusting it.

Good first use cases for a service desk:

  • Summarise a noisy 40-comment incident ticket into a three-line status
  • Answer how do I request VPN access by reading the actual IT runbook
  • Draft a customer reply in the right tone from the resolution notes

Cited answers let a human trust, then verify, before acting

Rovo Agents: From Answering to Doing

Agents are the leap from information to action. Atlassian ships built-in agents and lets you build custom agents with no code — you describe the agent's purpose, give it instructions, and grant it access to specific actions. Practical agents I would deploy:

  1. Triage agent — reads every new JSM ticket, classifies it, sets priority, and routes it to the right team
  2. Release-notes agent — collects all issues in a fix version and drafts a Confluence release-notes page
  3. Knowledge-gap agent — watches recurring tickets and suggests a new Confluence article when the same question repeats
  4. Standup agent — summarises what a team moved yesterday from Jira activity

Building a Custom Agent

A custom Rovo Agent is defined by instructions in plain language, not code. The pattern that works:

Name: JSM Triage Assistant
When: a new ticket is created in the IT Support project
Do:
  1. Read the summary and description
  2. Classify into: access, hardware, software, or network
  3. Set priority from the impact keywords
  4. Assign to the matching team and post a one-line internal note
Guardrails: never close a ticket, never email the customer directly

Notice the guardrails. The single most important rule when rolling out agents is to constrain their actions — read and suggest first, act autonomously only once you trust the behaviour.

Rolling Out Safely

AI that touches production tickets needs the same discipline as any deploy:

  • Start in suggest mode where the agent proposes and a human approves
  • Scope each agent to one project before going org-wide
  • Keep an audit trail — every agent action should be visible in the ticket history
  • Review agent decisions weekly and tune the instructions

An AI agent is an over-eager new hire with infinite energy and no common sense. Give it a narrow job, check its work, then widen its remit.

What to Learn Next

  • Rovo connectors to bring GitHub, Google Drive, and Figma into the graph
  • Confluence knowledge hygiene — agents are only as good as the pages they read
  • Automation plus Rovo — deterministic rules for the boring parts, agents for the judgement calls

Arivanandhan Chitheshwaran