Atlassian Intelligence: SaaS Co. Gets Generative AI Makeover

The Australian public company Atlassian has just launched a new generative AI feature in its cloud software suite. Called Atlassian Intelligence, it’s being touted as “a new virtual teammate” and was built in collaboration with OpenAI.
Ahead of the announcement, I spoke with Atlassian’s Sherif Mansour (the company’s Head of AI) and Luke Heinrich from its business operations department.
Atlassian Intelligence includes both internal models and technology supplied by OpenAI. It has been built into the underlying Atlassian platform, in much the same way as “collaboration” features are a core part of its platform too.
Atlassian’s two main products are Jira (project tracking) and Confluence (a team collaboration platform). It has other products too, but according to Mansour it all boils down to two use cases: project work and service work.
Teamwork Graph
The key to Atlassian’s new AI integration is a concept it calls a “teamwork graph.”
“What we’ve discovered over the 20 odd years,” Mansour said, referencing the amount of time that Atlassian has existed, “is that we’ve organically built this ‘teamwork graph’ of how teams connect and communicate, from project all the way to delivery, and then how they maintain that — whether it be a technical DevOps team, or even a non-technical team running a project such as a marketing campaign — and how they provide services to other teams in the company.”
“That’s really where we’re trying to unleash the potential of artificial intelligence, on that teamwork graph,” he added.
I asked when Atlassian came up with this concept, and Mansour replied that they refined it over the past 12 or so months, as a way to “think about how people connect and communicate” with its products. He also clarified what they mean by a graph.
“We use the term ‘graph’ to describe the collection of the APIs, and products and services we have, the data that we have, and the nouns that people have in our products — for example, a pull request in Bitbucket, an issue in Jira, a deployment in Pipelines.”
Atlassian Intelligence in Action
Like other enterprise software companies, Atlassian has had so-called “smart features” in its product line for a while now — “personalized search, relevance, all that kind of stuff,” said Mansour. He described Atlassian Intelligence as “really trying to bring generative AI to the forefront.”
Mansour then gave me a demo of the new functionality. He began by showing how generative AI technology from OpenAI is used to create, summarise and extract information from content within the Atlassian suite.
For example, Atlassian Intelligence can summarize the activity on a help request in Jira Service Management:
Another example is Atlassian Intelligence explaining an institutional term to a teammate reading a Confluence page:
In terms of more creative use cases, you could ask the AI to define test plans for product updates in Jira Software:
One of the more intriguing examples I saw was Atlassian Intelligence translating a natural language query to Jira Query Language (JQL), which is Atlassian’s SQL-like language for its products. This potentially means that Atlassian users will no longer have to upskill in JQL.
Percept.ai
Although all of the generative AI functionality in these examples comes from OpenAPI, Atlassian has also brought to the table its own AI technology. Primarily that was through an AI company it acquired in January 2022, called Percept.ai. That product pre-acquisition was a virtual agent, built using Percept.ai’s proprietary natural language AI technology.
Mansour explained that some of the examples he showed me use the Percept.ai large language model (LLM) mixed with ChatGPT technology. He described it as “a mix and match with different examples,” depending on the requirements for each use case.
As for its working relationship with OpenAI, he said that Atlassian has “dedicated infrastructure that’s isolated and a whole bunch of agreements with them.”
While Atlassian Intelligence can clearly do a lot with both OpenAI technology and its own internal Percept.ai models, there are limits to what AI can do for its users. For instance, if a developer is asking for the AI’s help to come up with a development plan for the latest iPhone models, they will have to bear in mind that OpenAI’s data models only go up to September 2021 at this point. So the AI won’t be able to tell the user about the most recent iPhone model.
Mansour and Heinrich told me this limitation in LLMs will need to be communicated to its customers, but they added that Atlassian Intelligence can also use external data sources in certain cases. An example they showed me was a marketer drafting a tweet based on product specs documented in Confluence.
How Will Enterprises React to Generative AI?
Judging by the examples I was shown, Atlassian Intelligence will be as just helpful to Atlassian users as ChatGPT is to the rest of us. With this release, Atlassian has also demonstrated that it’s a fast mover in the enterprise SaaS market. It likely will have a jump on its competitors with this generative AI makeover.
That all said, Atlassian will face the same questions about generative AI as the consumer market is asking. Can Atlassian’s customers trust the data (in other words, will the AI hallucinate)? Also, will their company data stay private if it’s being used by an AI system?
On the privacy front, Mansour insisted that everything he showed me will be opt-in by their customers.
“The vast majority of our customers are SMB and enterprise,” he said, “and they want high degrees of control. And so we have a central administration experience for all our products, and there are teams working on that right now to allow [you] to opt-in and opt-out.”
There will surely be other thorny challenges ahead for Atlassian Intelligence customers, but even so, this is an impressive integration of generative AI technology into enterprise software.