Setting Up Rovo AI Agents in Jira Without the Hype

Setting up AI agents in Jira means working through Rovo, Atlassian’s built-in AI layer. You can use one of the 20-plus prebuilt agents right away, or build your own from Rovo Chat, the Studio app, or a work item’s assignee picker. A custom agent gets a name, instructions, knowledge sources, and a few tools, and then you activate it.

That is the short version. The longer version matters more, because the gap between what Rovo agents promise and what they reliably do in a real Jira instance is where most teams waste their first month. This guide covers the setup mechanics, the credit and governance limits that quietly shape what is practical, and the cases where an agent is the wrong tool.

What Rovo is in 2026, and how agents fit

Rovo is now included in every paid Jira, Confluence, Jira Service Management, and Teamwork Collection cloud subscription. There is no separate Rovo SKU to buy for the core features. Agents are one part of Rovo, alongside Search, Chat, and the writing tools. Atlassian ships more than 20 agents out of the box and lets anyone build custom ones.

Two constraints are worth understanding before you touch anything. First, Rovo has to be enabled by an organization admin, and the organization must sit on a verified business domain. Atlassian will not turn Rovo on for hobby and personal-domain instances registered under generic addresses like gmail.com or outlook.com. Second, agent usage runs on Rovo credits, and the allowance is smaller than the marketing suggests.

Each request to a Rovo agent currently costs 10 credits. Jira Standard includes 25 credits per user per month, Premium 70, and Enterprise 150. Credits are pooled across your whole organization and reset monthly with no rollover. On Standard, that works out to roughly two or three agent prompts per user per month before you lean on the shared pool. Atlassian says it is not billing for overage today and will give at least 90 days’ notice plus an explicit opt-in before that changes, but the allowance is the number to plan around.

If you are still deciding which tier to sit on, our breakdown of Jira pricing for 2026 lays out what else changes between plans. The Free plan gets no agent access at all, so any serious Rovo evaluation starts at Standard.

Prebuilt agents versus building your own

Before you build anything, look at what already exists. Several out-of-the-box agents are aimed squarely at Jira work: the Work Item Organizer moves items into sprints and assigns them to epics, the Jira Theme Analyzer groups work items into themes from a JQL query, the Bug Report Assistant tidies up bug reports, the Readiness Checker tests items against a definition of ready, and the Blocker Checker is built to run inside automation rules. The full out-of-the-box agent list is worth reading before you reinvent one.

Be honest with yourself about which of these earns its keep. The analysis and drafting agents tend to save real time. Some of the lighter ones are thin wrappers around a prompt you could type into Chat yourself. Build a custom agent when you have specific knowledge to attach, such as your team’s Confluence space or a particular project, and a repeatable job that the generic agents handle awkwardly. Otherwise, duplicate an existing agent and point it at your own knowledge source rather than starting from scratch.

How to create a custom Rovo agent

There are three entry points. The fastest for a Jira-centric agent is the work item picker.

1. Open a work item in Jira.
2. Select the Assignee or Agent picker.
3. Select Create agent in the footer of the picker.
4. Rovo Chat opens with a draft plan. Ask it to adjust the plan if the scope is off.
5. Select Accept and assign to create the agent and assign it to that work item, or Refine in Studio to keep configuring.

For a reusable agent that is not tied to one item, build it from Chat or Studio and configure it manually. This gives you control over instructions, knowledge, and tools.

1. Open Studio from the app switcher, select Agents, then under Rovo agent select Create.
2. In the chat window, select Skip to manual setup.
3. Give the agent an Identity: a clear Name that signals its role, a Description that appears in the agent directory, and Behavior instructions it follows in every situation.
4. Add Conversation starters so people know how to prompt it. Leave this blank and Rovo supplies three generic ones.
5. Configure the default Subagents: write the Instructions, attach Knowledge such as a Confluence space or a Jira project, and pick Tools. Atlassian recommends no more than five tools to keep responses sharp.
6. Select Activate to publish.

The five-tool guidance is not a formality. Agents loaded with a dozen tools get slower and less predictable about which one to reach for. Start narrow, watch the responses, and add capability only when a real gap shows up. Atlassian’s own agent creation documentation reinforces the same point about iterating on instructions rather than front-loading everything.

Wiring an agent into automation

An agent on its own only responds when someone prompts it. To make it act without a human in the loop, it has to run inside an automation rule, and that is admin territory. There are two ways to connect the two.

1. Open Studio from the app switcher and select Automation.
2. Select the app and space for the rule, then Get started, which drops you into Automation for that space.
3. Select Create flow, then Create from scratch.
4. Add a trigger, such as work item created, and configure it.
5. Select Next, then THEN: Add an action, and choose Use Rovo agent.
6. Connect Rovo to automation so the agent can act on your behalf, select the agent, and write the prompt.
7. Add a second action, such as add comment or send Slack message, and pass the result through with {{agentResponse}}.

Alternatively, if you own or manage the agent, open it in Studio, select Triggers, and add an automation trigger from there. Either path lands you in the same Automation builder. If you are new to that builder, our guide to setting up Jira automation rules covers triggers, conditions, and the connection step in detail. A useful pattern: let a plain automation rule handle the trigger and routing, and hand only the judgment step, such as summarizing or classifying, to the agent. That keeps credit spend down and behavior predictable.

Admin controls and the permissions model

Rovo Studio is open to everyone in your organization by default, which is exactly how agent directories turn into a mess. A Studio admin can change that under Settings, choosing All users, Selected groups (up to 10 user groups), or No users, which limits creation to organization admins. If you are rolling Rovo out across more than a handful of teams, tighten this on day one.

The data model is the reassuring part. An agent always acts on behalf of the person invoking it and can only reach content that person already has permission to view or edit. If a user cannot see a page, the agent cannot summarize it. Sharing an agent never widens anyone’s underlying access. Ownership can be transferred, and you can add editors and managers per agent. One current gap worth noting for governance: you can restrict who sees an individual agent, but only by adding users one at a time, not by group or team.

Where Rovo agents underdeliver

The honest limitations are mostly about economics and scope, not capability. The credit ceiling bites first. At 10 credits per request and 25 per user each month on Standard, a team that actually adopts agents will drain the pooled allowance quickly, and heavy automation runs consume credits every time they fire. Model your expected volume before you build a rule that pings an agent on every new work item.

Scope is the second issue. Agents inherit garbage-in behavior. Point one at a stale Confluence space and it will confidently cite outdated decisions, because it has no idea the space is stale. The permission model that protects your data also means an agent is only as useful as the knowledge the invoking user can already reach. And autonomous behavior is gated behind admin-built automation, so the self-service demo you saw is not what a viewer-level user experiences on their own.

Skip agents entirely for anything with a fixed, deterministic outcome, where a plain automation rule is cheaper and more reliable, and for teams still on Free or a personal domain. If Jira’s overall weight is the real problem rather than the AI layer, a lighter tool may serve you better; our roundup of the best Jira alternatives for 2026 is a more productive place to start than forcing an agent onto a workflow that never needed one.

Frequently asked questions

Do I need a paid Jira plan to use Rovo agents?

Rovo credits are included with paid Jira, Confluence, and Jira Service Management plans, not the Free plan. Your organization admin also has to be on a verified business domain. Atlassian will not enable Rovo on sites registered with generic addresses like gmail.com or outlook.com, so Free-plan and personal-domain teams have no agent access.

How much do Rovo agents cost to run?

Each request to a Rovo agent currently costs 10 Rovo credits. Jira Standard includes 25 credits per user per month, Premium 70, and Enterprise 150, pooled across your organization and reset monthly. Atlassian is not billing for overage yet and has promised at least 90 days’ notice before that changes.

Can Rovo agents act without a human prompting them?

Yes, but only through automation. An agent cannot run on its own until a space or project admin wires it into an automation rule, either by adding a trigger in the agent’s configuration or by adding a Use Rovo agent action to a flow. Outside automation, agents respond only when prompted directly.

Can a Rovo agent see data users are not allowed to see?

No. An agent acts on behalf of whoever invokes it and can only touch pages, work items, and spaces that person already has permission to view or edit. If a user cannot see a page, the agent cannot either. Sharing an agent never widens anyone’s underlying access.

How are Rovo agents different from Jira automation?

Automation rules follow fixed if-this-then-that logic you configure in advance. Rovo agents add a reasoning step, so they can read a work item, weigh context, and draft a response. The strongest setups combine both: automation handles the trigger and routing, and the agent handles the judgment call in between.

Sponsored