What's New in Rovo Agents: Your Bi-Monthly Update
Curious about what the Rovo Agents team has been up to? You’re in the right place! Every two months, we’ll highlight the latest features, key updates, and give you a preview of what’s coming next.
April and May - at a Glance
7 Major Releases
It was a massive two months for Rovo Agents, with major updates to the builder, new automation capabilities, and better ways to manage your agent fleet.
3 minor releases
3 Minor Releases
We shipped improvements to the underlying AI models, added new knowledge sources, and made it easier to move agents between sites.
4 features coming soon
4 Coming Soon Features
We have some powerful updates in the works, including more control over how agents use tools and the ability to choose your own AI models.
🚀 Major Releases
Versioning and Drafts for Agents
We’ve added versioning and drafts so you can work on your agents without worrying about breaking anything for your users.
Now, you can build and test a draft version of an agent before you hit publish. You can also see a history of your changes and go back to an older version if you need to. This is great for agents you use in live demos or production where you need things to stay reliable.
Building Smarter with Subagents
We’ve refreshed the Studio experience to make it easier to build powerful agents using reusable building blocks. This update includes a modernized builder interface and support for subagents. Instead of building one giant, complex agent, you can now connect smaller, specialized agents together to handle more varied tasks.
If your agent uses subagents, you can now choose exactly how the main agent handles their responses:
-
The agent can modify the output (Default): The main agent collects information from multiple subagents and combines it into one final answer. This is best for complex workflows where you need to pull data from different sources, like searching across both Jira and Confluence at once.
-
The agent must use the exact output: The main agent picks the most relevant subagent and passes its response directly to you without changing a word. This is ideal for service desks or help centers where specific areas (like IT or HR) have their own strict formatting or specialized instructions.
We’ve also updated the debugger so you can see exactly which subagent was called and which tools it used. This makes it much easier to track how your agents are thinking and where they’re getting their information.
Better Control with Knowledge Filters
You can now get much more specific about the information your agents use by adding custom filters to your knowledge sources. Previously, you could select a source, but you couldn't narrow it down. Now, an "edit" button on each source lets you set specific rules based on the type of data.
For example:
-
Slack and Teams: Filter by specific channels or only include conversations where certain people were mentioned.
-
Salesforce: Use filters specific to your CRM, like narrowing results down to certain deals or accounts.
-
Google Drive or SharePoint: Limit the agent to specific folders or file types.
These filters work behind the scenes to keep your agent focused. If you set an agent to only look at project updates from your specific team, it will follow that rule even if you don't mention it in your chat. This makes it much easier to build agents that are tailored to specific products, integrations, or team workflows without them getting distracted by irrelevant data.
Agents in Automations (Early Access)
We’re starting an Early Access Program to help you build more autonomous workflows. Rovo Agents can now act as "workers" within your Automation rules, taking over repetitive and routine tasks so you don't have to. We’re keeping this limited to a small group for now so we can gather feedback and make sure everything is running smoothly.
Agent Accounts and Access (Early Access)
To give you more control over how agents work on your behalf, we’ve launched an Early Access Program for Agent Accounts. This update creates distinct accounts and access levels for your agents. It’s a big step toward better governance, as it lets you define exactly what an agent can and can't do when it's acting for a user or a team.
Live Conversation Review (Closed Beta)
We’re opening a closed Beta for live conversation review. This tool lets agent creators and managers see the conversations people are having with their agents. It’s designed to help you monitor performance, tune responses, and ensure everything stays compliant. Right now, the program only supports public Slack conversations, but we’re working on expanding this soon.
If you’d like to join the Beta, please https://privacy.atlassian.net/servicedesk/customer/portal/698/create/3000 to start the onboarding process.
Top Trending Agents Dashboard
We’re rolling out a new dashboard for Enterprise and Premium customers to help org and studio admins see which agents are gaining traction. This will be generally available in early June.
The dashboard shows you the most popular agents and how their usage is trending over time. It’s a simple way to spot "breakout" agents that are performing well, so you can decide which ones to promote, invest in, or showcase to the rest of your team.
🛠️ Minor Releases
Faster, Smarter Models for Rovo Agents
Over the last two months, we’ve upgraded the AI models that power Rovo Agents to make them faster and more reliable.
-
Main Workflow: We’ve moved from GPT-4 to Google’s Gemini 3 Flash. This change keeps the quality of responses high while significantly reducing wait times (latency) for tasks like summarizing documents or planning steps.
-
Helper Tasks: For lighter work, like generating action messages, we’ve moved from GPT-4 mini to GPT-5 mini.
What this means for you:
Most users will notice snappier, more consistent responses, especially when agents are searching for and retrieving information. These updates are already live, so your existing agents are already benefiting from the improved speed. No action is needed on your part, but if you notice any issues with a specific agent, please reach out so we can look into it.
Agent Import / Export
We’ve added the ability to export and import your agents as JSON files, making it much easier to move them between different sites and environments.
This is a huge help for teams who want to test an agent in a sandbox environment before moving it into production. Instead of manually recreating your instructions and tools on a new site, you can simply export the file and upload it where you need it. This ensures your agents stay consistent and saves you from repetitive setup work.
Bitbucket and Loom Knowledge Source Support
You can now connect Bitbucket and Loom directly to your agents as knowledge sources. This means your agents can pull information from your code repositories and video transcripts to give more accurate and context-aware answers. Whether you're looking for technical details in a repo or a specific decision made during a recorded meeting, your agents now have a much wider range of information to work with.
🔮 Coming Soon: Sneak Peeks
From Skills to Tools
We are starting to transition the way we talk about agent actions. Soon, you’ll see "Skills" renamed to "Tools" across the platform. This isn't just a name change; it’s the first step toward a more modular way of building agents. By treating actions as tools, we’re making it easier for you to mix, match, and reuse different capabilities across your entire agent library.
Better Control Over Agent Tools
We’ve added new settings that give you more precise control over how your agents use their tools. These improvements make agents more flexible by allowing you to define exactly how and when an agent should trigger a specific action. It’s a great way to fine-tune your agent's behavior and ensure it handles tasks exactly the way you want.
Choosing the Right Model for the Job
We’re starting to roll out a new feature that lets you choose the specific AI model and "reasoning tier" for your agents. This is currently available to Atlassian staff and a small group of early access customers.
The goal is to let you decide which model works best for your specific needs. For example, you might choose a faster, lighter model for simple tasks to keep things snappy, or a more powerful model for complex work where accuracy is the top priority. This flexibility helps you find the right balance between speed, quality, and cost.
More Powerful Problem Solving with Marathon Agents
For tasks that require complex data analysis or technical execution, we’re introducing the "Marathon" orchestrator. Unlike standard agents, Marathon agents can write and run Python code in a secure sandbox to solve problems. This allows the agent to perform sophisticated calculations and return results grounded in actual code execution.
What’s possible:
In a recent internal demo, a Marathon agent took a single link to a technical blog post and—without any extra tools or specific instructions—read the content, wrote a script, and generated a complete 30-second Instagram Reel video. It even handled the technical side by installing the necessary libraries and procedural code to build the video from scratch.
💬 We Want to Hear From You
These features are built based on your feedback. Have ideas for what's next? Questions about how to implement these new capabilities? Reach out to your Atlassian team or visit our documentation to learn more.
Stay tuned for next bi-monthly update - we've got even more exciting features in the pipeline!
Was this helpful?
Thanks!
Rachelle Rathbone
2 comments