Atlassian Team ‘26 Europe: how AI becomes a multi-player sport

We spent three days in Amsterdam at Atlassian Team ‘26 Europe looking at how software delivery teams can scale AI without creating dark silos. Last year, the focus was on isolated apps; this year, Atlassian showed how humans and agents deliver together across the software development lifecycle (SDLC).

We tracked three main themes: making AI work visible to the whole team, shortening loops from idea to spec, and preventing operational incidents.

AI work becomes visible to the whole team

Atlassian introduced the Agentic Multiplayer Protocol (AMP). It is less a feature than a way of working, supported by new product capabilities across the suite.

The core premise is straightforward: when an engineer works with an AI chat assistant in isolation, it creates a private interaction between one human and the agents they are triggering. AMP makes the work of agents and humans visible to the rest of the team so collaboration remains central.

Here is what that looks like in practice:

• Shared chat sessions: Share a useful Rovo chat with your colleagues, so nobody has to copy pieces out, rewrite the context or send screenshots of the answer.

• Kanban agent activity: Cards display live indicators when an autonomous agent is working on a task, showing exactly what it is doing.

• Unlinked work detection: System scans surface unlinked side quests and spikes, allowing you to link them to work and create traceable Jira tickets.

• Explicit human-agent attribution: Audit logs record not just that an agent did something, but that an agent invoked by a named person executed it.

That last point matters deeply to us. Many organizations want agents to do more, but they also need strict governance. The agent executed the task, but it did so on your say-so. The ultimate responsibility stays with the human.

Software firms found that adopting AI coding agents raised lines of code by 30% and pull requests by 23%, but finished work did not rise significantly. Code review was the bottleneck, with time from pull request to merge up 49%.

According to Atlassian's official announcement, AMP addresses this directly by standardizing context sharing between agents to eliminate duplicate work and keep reviews moving smoothly.

Scaling AI is not about making individuals faster in a vacuum. It is about making AI work shared, traceable, and governed.

Planner and spec-driven development

Planner is now in early access. You start by describing a brief. Through context-aware follow-up questions, Planner helps you break it down into a technical specification and user stories, so you can start orchestrating the work.

Atlassian demonstrated this in a Loom, where talking through your design out loud feels natural.

Spec-driven development can sound like a return to waterfall. Here, however, the feedback loop takes minutes rather than weeks, making experimentation cheap. If the spec is wrong, scrap it and try something different.

The hard part has moved. Writing raw specifications, thinking, and setting strategy are where true engineering value sits today. Describe the outcome you expect and how you will know it is right. We should do this for people, and we need to double down on it for AI.

Planner takes on the administrative heavy lifting, while your team verifies that the result fits your broader product direction.

Rovo and Jira Service Management: preventing incidents

In Jira Service Management (JSM), three platform capabilities stand out for improving service operations:

1. Work where you work: You can resolve incidents and handle mitigation tasks directly in tools like Microsoft Teams and Slack without switching context.

2. Proactive prevention: An early access program brings Atlassian Rovo to local client devices to identify potential problems before they trigger incidents.

3. AI risk assessments: Available today, these assess pull requests and system changes to help you understand impacted stakeholders, compliance and business risks, testing coverage, and rollback plans.

Atlassian also demonstrated a solution composer where you talk to Rovo to set up service desks, workflows, automation rules, SLAs, and queues. It speeds up initial configuration, though setup time was rarely the main bottleneck in service delivery.

Solid operational foundations determine your final outcomes: connected services, clear ownership, and a smooth flow from idea to value. AI speeds up a good platform; it does not replace one.

Smaller launches that add up

Several AI focussed features to improve efficiency were announced to resolve daily developer friction:

• AutoDev: Takes on small bug fixes and feature requests. You assign a credit budget and specify which type of tasks the agent should take, allowing developers to keep their focus on higher-value work.

• Rovo Desktop for Mac: Redesigned with a compact overlay powered by the Teamwork Graph, enabling developers to pull company-wide context without changing context windows.

• Loom to agent: Lets creators record a screen share describing a desired workflow assistant; Loom generates the agent configuration automatically to solve the "blank page" problem.

• Rovo templates: Ready-made use cases you can pick from, so you don’t need a long list of out-of-the-box agents. 

What to take with you

• AI works best as a team activity: Visibility and human accountability must be built in from day one.

• Strategy and specification matter most: Experiments are cheap, so the quality of your specification and strategy matters more than ever.

• Fix your base platform first: High speed on top of a weak operating model just automates the mess.

Want to turn these announcements into a clear roadmap for your team? Explore our AI in the SDLC assessment and How to get more out of Rovo or get in touch with our team to discuss your platform journey.