Atlassian outsources office drudgery to GenAI agents

Atlassian is the latest vendor to have a crack at addressing the perennial problem of enterprises scattering data across multiple applications and then finding it's impossible to understand what's happening and why.

The Australian developer's attempt is called Rovo and, as explained to The Register by distinguished product manager and head of product for Atlassian Intelligence Sherif Mansour, it creates a "teamwork graph" that allows generative AI to summarize the state of a project.

Mansour said building the teamwork graph can start by identifying things like the presence of a Figma link in Jira and Confluence – an obvious sign of a relationship. Atlassian is building connectors, so that Rovo can find relationships in data stored across thousands of SaaS services, and even on-prem apps – like SharePoint, which was itself touted as a fix for enterprise data strewn across different applications.

The result is a pool of data that can be queried in natural language. Mansour used an example of asking "Who is working on Blueberry" – in which "Blueberry" is an internal codename for a project. Rovo will then return a summary of the project, perhaps listing its goals or providing a list of team members, or links to key documents like a launch plan.

Rovo will be accompanied by "Rovo agents" – software that Mansour described as "virtual teammates" that do low-value scutwork.

His example of an agent was called "Backlog Buddy" – a Jira agent that looks at a product spec, analyzes progress, then suggests a list of recommended next steps to advance a development push. That's a job Mansour explained is mostly low-value and nobody enjoys, and therefore ripe for offload to AI.

• Atlassian loses half its CEOs, but customers stay solid after Server products exit support
• Atlassian Confluence Server RCE attacks underway from 600+ IPs
• Atlassian buys 'asynchronous video' outfit Loom for almost $1 billion
• Atlassian pipes software flaw reports into Jira, so the boss can see them too

Another agent Mansour described monitors for mentions of new issues then helps resolve them. He outlined how the agent could detect an issue that calls for creation or revision of a graphic, synthesizes internal chat about it, considers the content of a company's brand guidelines (which were sucked into the Team Graph ages ago), and squirts info over an API to SaaS-y graphics service Canva, which would use generative AI to produce a candidate image.

That image would appear in the group chat, where workers banter about it. If selected as appropriate, the image is delivered and Canva gets paid.

Rovo is being tested by several thousand Atlassian customers, with Mansour "gunning to go live in late 2024." Pricing is yet to be set, but will be per-seat like most Atlassian products.

"We think the future of work is teams controlling multiple agents," Mansour told The Register. Atlassian already has hundreds of them at work – and Mansour expects that one day you will too.

That's a prediction that brings to mind other deep-seated tech problems, like shadow IT and sprawl. The Register can easily imagine that a few years from now, Atlassian will advise customers it's time to stop letting everyone create Rovo agents – because it's possible to have too much of a good thing. ®