Rovo is Atlassian’s AI surface. Custom agents sound inevitable; many teams only need better Automation and cleaner data. Here is a sober way to decide whether a custom Rovo agent (sometimes with Forge middleware) is worth it.
Start with a job story
“When an L2 agent opens a P1, the agent should summarise linked Confluence runbooks and suggest the Assets object”—that is a job. “We want AI in Jira”—that is not.
Prefer configuration when
- Native Rovo skills already cover search/summarise for your content.
- The risk of hallucinated actions outweighs the time saved.
- Your Assets/Jira permissions are messy (AI will amplify the mess).
Build a custom agent when
- You must bind proprietary tools (internal APIs) as skills with strict auth.
- You need a fixed skill set so routing does not drift to the wrong tool (a common Assets 401 class of failure).
- Marketplace or internal distribution needs a Forge connector / install lifecycle.
Engineering guardrails
- Human-in-the-loop for write actions (create issue, transition, comment).
- Audit every tool call with actor, inputs, outputs.
- Red-team prompts for prompt injection via issue descriptions.
- Document data leaving the Atlassian boundary.
Cheaper alternatives to try first
- Forge Automation action for deterministic enrichment.
- Issue panel that shows the three facts agents always ask for.
- Better request types and required fields so humans (and AI) get cleaner input.
Need this built? HostHob ships production integrations and Marketplace-ready apps from Rotterdam. See our services or Explore a Rovo / Forge pilot.