Atlassian Rovo

A lot of the announcements at Team24 revolved around AI and how it will be integrated with all of Atlassian’s various tools. One big announcement, however, was about a specific AI feature called Rovo.

Rovo helps teams out by performing in three areas - search, learn and act.

Rovo Search

Companies have data, lots of data… how much data, you ask? Well, on average it’s something something like 2 billion data points spread across 200 systems.  There is absolutely no way any sane person could make sense of that. Personally I have enough challenges keeping track of the information in the few systems I happen to use regularly that I don’t even consider how it can be combined or used with data from other places. 

So we are forced to use tools.  Unfortunately many of those tools only see a small portion of that data, making them useful only in specific instances.  If I’m very lucky I’ll have an analyst or someone else who is familiar with other data sets that can help me make sense of everything - but that’s uncommon. Even when I do have an analyst who can help me out, we’re still frequently unable to pull everything together - either due to technical limitations (a specific data set isn’t available to us) or due to access or other considerations.

Rovo however, overcomes these challenges. It natively plugs into every Atlassian product, and knows how to interpret that information. This alone is a massive advantage over a human… many times I have to teach someone what data is available, what it means, etc… whereas Rovo knows all of that out of the box.

Even better, Rovo can be connected to non-Atlassian systems to gather information. Sure, this takes a bit of setup, but once it’s done, Rovo is able to analyst and add that data to what it already knows. While a human can certainly do this it takes time to learn the data sets (just like someone unfamiliar with how a Jira ticket is structured has to learn it), which slows down the process and opens it up to risk.

There is one big question that needs to be answered… Security. In order for Rovo to be tied into various systems, it inherently must have access to them. This means if I use Rovo to help me, it’s possible I’ll be exposed to data I don’t have access to… or does it?

It does not! Rovo is designed to respect the permission schemes of the source systems - so if I can’t access it with my credentials, Rovo won’t show it to me (even if Rovo can see it).

Rovo Learn

Data is great - it helps us make decisions, discover new things and make progress. Raw information, however, is generally useless without some type of analysis or organization. For example if you see a project name in a Confluence page it doesn’t really help you if I show you every page that mentions it. Humans are pretty good at finding patterns, discerning meaning and understanding what's going on, however, with the massive amounts of data flying around this is limited.

We are, after all, limited in the amount of information we can keep in our head. (I can barely remember where I leave my keys some days…) Systems like Rovo don’t have this limitation. (Sure there are technical limitations on it, but for the purposes of this discussion they’re effectively meaningless compared to the limitations a single human runs into). Rovo uses this capacity to help humans make sense of all that data.

For example, Rovo can assemble knowledge into “knowledge cards” that display information about a project including team members, milestones, metrics and more. This instantly allows a human to understand more about a project or team - and doesn’t require that human to pull apart multiple Confluence pages or other sources to learn that. Rovo does this by having access to all that underlying data, and is able to pull together contextual information from multiple places. 

Rovo can also define terms and acronyms. Personally I’m a huge fan of guess what TLA (Three Letter Acronyms) mean, but not know, or not being able to easily figure them out, can be incredibly frustrating to folks joining a team. Rovo is able to do this across Atlassian, and non-Atlassian, systems. So not only will it define terms you see on a Jira ticket, but also in a Google Doc, or Outlook Email. Imagine how much less frustration folks would have with this type of information easily accessible.

And last but not least, Rovo supports natural language chat. This can either be done from the get-go, or as followup questions about anything it serves up. For me this is a massive plus as it allows users to gain trust in the system by asking clarifying questions. Rovo will even show where it found its information, giving users more confidence that the system is giving them positive information and not hallucinating or giving incorrect info.

Rovo act

Being able to parse through data and determine context, meaning and other things is a great ability of Rovo… however, it does take it one step further and helps users take various actions based on the data it see. Rovo has the concept of “agents” - essentially virtual team members - who can be designed to perform specific tasks.

Agents can be shaped to assist humans in a variety of ways.Imagine waking up and seeing a list of suggestions for how to prioritize your backlog, based on information Rovo knows about the project, its status, and team members. Or what if you needed to draft a marketing document, just ask a Rovo Agent - which has been developed and trained by your own marketing team on your company’s assets and tone - to help.

These Agents appear in the same places human team members appear, making them feel like a team member. To me this blurs the line between “tool” and “team member”, but not necessarily in a bad way. After all, learning how to use a tool can take time and can be frustrating, but if I treat that tool like a virtual team member, it cuts out a lot of the learning curve and friction.

Rovo Agents can be built from raw code, allowing developers a great deal of flexibility in defining what they do, how they behave and what access they have. They can, however, also be developed via no-code options. This allows less technically-inclined teams to develop and share agents with other teams - further enhancing their impact and reducing friction.

My take

We're only getting more systems and more data, so having a tool that is plugged into as much of it as possible and can help us make sense of it all is an obvious next step. This is double true when considering the amount of manual effort it would take to approximate this capability.

I can, however, see challenges in gaining adoption. For example groups may question how secure it is, or wonder how they know they can trust the tools output. After all, many LLM’s and AI operate in a bit of a black box, which can make it hard for humans to trust them (especially if they are used to being able to ask an analyst or other expert how they got to their conclusion.

These concerns, however, aren’t a reason to not develop or use this tech… it is, though, a reminder to be careful in how these tools are introduced to organizations and how they are managed. For example, instead of a “big bang” rollout begin small and find a group that's excited about it and develop targeted use cases. Work with them to determine other use cases, potential issues and other areas to expand into. Then let that group tell others about how useful it can be.

Overall I'm excited for tools like Rovo to come out, they’ll help free us up to do other more interesting things than fight with data. They’ll not only reduce the monotony that exists when wrangling large amounts of data, but also open new areas of inquiry and illuminate areas that otherwise would have been hidden. I can also easily see this tech help smaller teams have an outside impact by allowing them to extend into other areas.