Atlassian says its new protocol fosters human-agent teamwork more effectively than headless or control-tower approaches. Analysts say data and trust are the true differentiators.
Atlassian launched a series of updates to its agentic products this week, betting that governable, human-AI workflows with shared enterprise context will get users working alongside and trusting AI agents as they would a colleague. But the key for customers will be whether they are already trusting Atlassian to store and curate context data for agents, according to analysts.
The company added Rovo Work for long-running tasks, Rovo Code Search, Artifacts for sharing AI-generated work, and the Agent Multiplayer Protocol to enable agents to draw on Atlassian's Teamwork Graph. These features give developers a window into what the agents are doing in real-time and allow them to work from a shared context.
Atlassian's claim is that what it calls 'multi-player mode' is a preferable approach to letting agents roam free in headless mode or dictate what gets escalated to managers through a separate governance app.
"There's some wisdom out there right now around, 'Oh, let's just go make your platform headless and that'll solve all problems,' Valliani said. "Headless is brainless … I don't think any customer is paying us to give them an MCP interface, then just walk away and pretend we don't have to help you get your agents and team members operating together."
Valliani also alluded to a 'Control Tower' approach as being too separated from day-to-day workflows.
"There are a lot of companies out there that are saying, 'Ah, we give you a control tower, now go away,'" he said.
Valliani didn't specifically name Salesforce or ServiceNow, but the terms 'headless' and 'control tower' are closely associated with those ITSM rivals' products. Salesforce said at Dreamforce this year that it would expand its Headless 360 offering, betting that headless would become the new interface for enterprise agents, despite its years of investment in its user interface. ServiceNow's Autonomous Workforce, launched earlier this year, is governed by a product called AI Control Tower.
For agents, context is key
Atlassian is one of a slew of vendors arguing that their platforms provide the context agents need to succeed, said Torsten Volk, an analyst at Omdia, a division of Informa TechTarget.
"Microsoft, ServiceNow, Salesforce and GitHub are making the same argument with their own graphs, so the differentiation for customers will come down to which graph already holds most of their work, and Atlassian's advantage there is Jira and Confluence, not the AI," Volk said.
Torsten Volk
Thus, while Atlassian officials took swipes at the headless Salesforce approach to agent collaboration and at ServiceNow's Control Tower, Atlassian's announcements will resonate most with its existing customers, said Rebecca Wettemann, CEO and principal analyst at Valoir.
"If I have a longstanding relationship with Salesforce or ServiceNow, and I'm comfortable with that, I can trust them, I can trust their technology, then I'm going to go with them. I think it's going to be hard in the long run for an emerging player to gain the kind of trust that they would need for people to make big bets," Wettemann said. "If I take a bigger step back and look at the core model guys, would I trust an OpenAI [to do orchestration] today? I don't know."
The differences in performance between frontier models are shrinking and increasingly hard to measure for most of the work that happens inside the software lifecycle, so what sets one agent apart from another is context and governance, Volk said.
"The Teamwork Graph, not the model, is what [Atlassian] is positioning as the differentiator, [which, along with] the new capabilities, extend what the graph captures, from source code at the function level to warehouse data, third-party conversations and a record of what every agent did and produced," he said.
The Teamwork Graph, not the model, is what [Atlassian] is positioning as the differentiator, [which, along with] the new capabilities, extend what the graph captures.
Torsten VolkOmdia Analyst
Over the past year, Atlassian has steadily expanded the context available to AI agents, adding structured enterprise data to its Teamwork Graph, opening that graph to third-party agents through MCP and bringing AI agents deeper into software development workflows.
In this update, Atlassian focused on enriching its existing Teamwork Graph with the technical and business context that agents need to carry out tasks. Two examples of that are Code Search, which gives agents access to source code, and Secoda connectors to data warehouses such as BigQuery, Snowflake and Databricks. A new Artifacts app stores, shares and controls access to outputs created by Rovo Work and other agents.
"Right now, most of these things are like little flat files and people just start attaching them and sending them off everywhere, so the permissions wind up going out the door," Valliani said. "If we create a proper place where artifacts can be stored and shared and easily accessed across agents, then we can make that both more efficient and also a lot more secure."
Jamil Valliani
A record of what every agent produced, combined with the source code and warehouse data added to the Teamwork graph, could foster more accurate outputs from agents by providing a much smaller space for a model to guess in, Volk said.
"An agent asked to fix a checkout service that slowed down after last week's release would not have to reason from 'anything that could slow down a service," he said. "It could find the function that changed in that release, query the warehouse for the drop in conversion rate since the deploy, pull the recording where the team discussed the tradeoff behind the change, and read the session log of the agent that made it. That is one function, one release, one documented tradeoff, and one prior agent run, each of which it can cite rather than infer."
Having a record of agent actions might allay fears about agents going rogue and causing damage without companies being aware of it, Wettemann said.
"People are afraid that they're going to wake up in the morning and AI is going to have deleted their database or sent a bunch of sales offers or done something else deleterious to their business. Or they're going to wake up and have a $1,000,000 token bill," she said.
Ben Lutkevich is an award-winning writer for TechTarget.
Torsten Volk
Jamil Valliani
