Atlassian Team ’26 Europe Amsterdam: Recap and Key Takeaways

In 1893, a group of Dutch bicycle makers founded a trade association. They later added cars, called it RAI, and built a convention center. On October 6 to 8, 2026, that same building hosted a conference about the newest member of the team: the AI agent.

Atlassian Team ’26 Europe brought thousands of people to Amsterdam RAI for three days of sessions, keynotes, demos and hallway debates. The headline was clear from the founder keynote on Wednesday morning. Atlassian wants AI to stop being a private chat window and start being a teammate, with a name, an owner and a seat at the board. Mike Cannon-Brookes put it in one line: “Agents aren’t replacing people, they’re joining teams.”

For three days, our team ran live demos at booth 407 and talked files and workflows with attendees, customers, partners, and Atlassian Community Champions. Between demos, we sat down with four of those champions and asked them the questions we think matter most right now. Their answers run through this recap, alongside what Atlassian announced and what we think it means for your teams.

At a glance

• Where and when: Amsterdam RAI, October 6 to 8, 2026
• Big announcement: AMP, the Agentic Multiplayer Protocol, which sets the rules for agents working alongside people
• Big number: the Teamwork Graph now connects more than 250 billion objects and relationships
• Big shift for Europe: local EU AI inference, which keeps LLM processing on models hosted in the EU
• Big question: 85% of knowledge workers use AI, but only 6% of executives see clear organization-wide ROI
• Next stops: Team ’27 in San Francisco (April 20 to 22, 2027) and Team ’27 Europe in Copenhagen (October 5 to 7, 2027)

Atlassian Team Europe 2026 in Amsterdam, October 6 to 8

Atlassian Team Europe 2026 took place at Amsterdam RAI, in the Zuidas business district. The venue is no stranger to big crowds. It hosted the 1970 Eurovision Song Contest and welcomes up to 2 million visitors a year across 11 halls and 22 conference rooms.

The format followed the Team recipe: the founder keynote on Wednesday morning, October 7, product deep dives, an Expo Hall full of Marketplace and Solution Partners, and evenings that turned into community meetups. For those who couldn’t make it, Atlassian put the keynotes and featured sessions on Team ’26 Europe Digital, including the closing keynote from Thursday afternoon, October 8, which is available on demand until November 7, 2026.

From Team ’26 to Team ’26 Europe: the end of single-player AI

Team ’26 Europe was not a replay of Anaheim. In May, Atlassian Team ’26 opened up the Teamwork Graph and gave agents a place in Jira. Five months later in Amsterdam, it announced the rules for how those agents behave once they’re on the team.

What AMP actually is

Atlassian announced AMP, the Agentic Multiplayer Protocol, at Team ’26 Europe on October 7. In Mike Cannon-Brookes’ words, most AI today is “single-player”: one person, one chat window. AMP is Atlassian’s attempt to make it multiplayer. It works in three layers:

1. In-the-flow collaboration. You brief an agent with a Loom video, @mention it in Confluence, or bring it into a Jira comment thread. Through the Atlassian MCP Server, it can also pull context from tools like Figma and your IDE.
2. Identity and shared presence. Every agent has an owner, a profile, and real-time attribution. Agents show up in presence bars and cursors on whiteboards, just like a colleague.
3. Context and governance. Agents draw on the Teamwork Graph, run either “as the user” or with a dedicated service account, and leave a full audit trail.

Atlassian says AMP is “live across the Atlassian platform today,” and that people and agents already work together more than 10 million times a month.

Peter Van de Voorde, who has spent more than 19 years in the Atlassian ecosystem advising Marketplace and Solution Partners, summed up the shift for us at our booth: “You still need to think about teamwork, but now include agents and humans, all focusing on building the most value for your customer in a structured, enterprise-ready way.”

Teamwork Graph: why context was the word of the event

One word ran through almost every announcement: context. The Teamwork Graph is Atlassian’s context engine: a living map of work, code, people, goals and decisions across Jira, Confluence, Loom and connected apps. It now holds more than 250 billion objects and relationships, up from 150 billion connections in May, with 80+ connectors including new ones for Zoom, Gong, Microsoft Entra ID and Google Identity.

Why does it matter? Because agents without context guess. Atlassian’s engineering blog says it best: “AI agents need good context, or they get very confident about the wrong thing.” In Atlassian’s internal testing, agents grounded in the Teamwork Graph gave answers up to 44% better while using up to 48% fewer tokens.

The pieces announced around it:

• A rebuilt Atlassian MCP Server with 200+ tools, up from dozens, using up to 25% fewer tokens for the same Jira and Confluence work.
• Rovo Code Search, so you can ask questions about code in Bitbucket and GitHub in plain language, without cloning a repo.
• Atlassian Insights, which brings structured data from data lakes and warehouses into the graph.
• Third-Party Folders in Confluence, which sync content from apps like Google Drive into Confluence folders, so a sales deck stays current without copy and paste.

That last one caught our eye. Atlassian is openly saying what customers have told us for years: a lot of company knowledge doesn’t live in Jira or Confluence. It lives in SharePoint, Google Drive, and other cloud storage.

We put that exact point to Peter Van de Voorde. His answer: “You don’t need the entire internet, but you do need the context that the company needs.” In big companies, he added, you might not even know which tools other teams use. “You need to go on a journey of discovery: find the places where all that context and content is, and bring it into the fold.”

The other champions said the work starts before any connector is switched on:

• Anahit Sukiasyan, Community Champion for Yerevan and a Jira administrator at a bank: keep everything documented, “not only word-of-mouth,” with processes written down and tickets for the work done.
• Nikola Perišić, Atlassian platform owner and Community Champion, pointed at the other half of the equation: the prompt. Tell the AI what output you need, what to emphasize, what to avoid, and what role it plays.

What was announced across the five Collections

Atlassian groups its apps into five Collections built around outcomes. The short version: agents are moving into every one of them, and Atlassian now wants to measure what they’re worth.

The numbers worth stealing for your next leadership meeting come from Atlassian’s AI SDLC transformation playbook. In Atlassian’s 2026 study, 94% of engineering leaders say their teams use AI, but only 6% have the systems to scale and govern it. Inside Atlassian, AI-native development cut issue cycle time from 11.5 to 6.7 days, and agent loops in Jira automatically resolve 51% of potential vulnerabilities. Its sharpest line is a warning: leaders who treat AI “primarily as a procurement exercise” will see “incremental gains at best.”

The State of Product 2027 report, based on 1,000 product professionals, adds an uncomfortable twist: “Product managers say AI helps their teams ship faster, but customers aren’t seeing value any sooner.” If delivery speeds up but value doesn’t, the bottleneck has moved upstream, to deciding what to build.

DX was the favorite of two of our champions. Deepanshu Natani, who manages the Atlassian consulting team at Modus Create, said the question has changed: “Till last year, everyone was just thinking that we need to start using AI, but now the question has come: what is the ROI?” Nikola Perišić called DX the best thing he saw at the event, pointing to the demo of token usage by model and “a price per pull request.”

Trust, governance and the European angle

Amsterdam was a fitting place to talk about guardrails. The Dutch measure height against NAP, the Normaal Amsterdams Peil, which roughly matches average sea level, and almost half of the Netherlands lies below it (NAP). A country that keeps the sea out with careful water management knows that the rules around a powerful force matter as much as the force itself.

For European teams, the most important announcement may not have been the flashiest. Atlassian announced local EU AI inference, which “restricts LLM processing exclusively to models hosted within the EU.” No availability date was given. Jamil Valliani, Atlassian’s Head of Product for AI, told diginomica that it meant negotiating capacity with hyperscalers: “We’ve approached this as a fundamental infrastructure investment.”

Why it matters: today, by default, Rovo routes requests to a mix of Atlassian-hosted and third-party LLM providers. Cloud Enterprise customers can request Atlassian-hosted models only (Atlassian Trust Center). EU inference adds a regional guarantee on top.

Around it, Atlassian announced a wider governance package: agent accounts and a non-human identity inventory (coming soon) so admins can see and revoke every agent, app, and service account; Guard Premium for finding and controlling sensitive data, including attachments; and proactive Rovo compliance for the EU AI Act and ISO 42001, backed by audit logging.

What should admins check before letting agents in?

The answers depended on where each champion sits.

• Nikola Perišić starts with permissions: can the agent read, create, or only view, and which data does it touch? For banks and financial companies, he pointed to Guard Premium policies that block requests containing sensitive data like account numbers. He also reminded admins that many people skip Rovo and connect ChatGPT, Gemini or Claude through MCP, so those connections need the same controls.
• Peter Van de Voorde says it depends on who you are. A three-person startup building an MVP? “Don’t worry too much about it; just go and see what happens.” A defense company with 20,000 employees and government contracts may want to run as much as possible on its own infrastructure, “keeping it completely air-gapped even, sometimes.”
• Anahit Sukiasyan asks whether the organization is ready at all: is the documentation, process, compliance and governance in place? Close those gaps with the teams first, then adopt AI.
• Deepanshu Natani sees trust growing as vendors make clear they don’t train models on customer data, but expects teams in constrained industries to move more slowly.

Our take: EU inference and agent identities answer the questions Nikola and Peter raised. The question is how quickly they ship, and whether teams will use them before they need them.

The champions’ debate: the admin job and Data Center’s last years

Two more topics came up with every champion: what AI does to the Atlassian admin role, and what to do before Data Center reaches end of life on March 28, 2029.

Will AI change the Atlassian admin job?

Three champions, three different answers.

Peter Van de Voorde: admins become change managers. “They need to make that change manageable for all the users,” he told us, teaching people to work with agents, MCP, and the new apps “while also understanding the nuances and culture of the company.” He called it “a completely new skill set,” and a hard one for technical admins.

Deepanshu Natani: technology is only 30% of the job. His example was the most memorable line we heard in Amsterdam. A project manager asks for a new custom field. Any admin can create it in two minutes, and so can AI. But a good admin first asks whether an existing field already does the job. “The thing with AI is, AI never says no, at least till now,” he said. Multiply that by 700 projects with 700 workflows, and the people working across them get lost. “The human part, the process part, the people part, that’s 70%.”

Nikola Perišić: concerned, but not worried about replacement. He sees more admin tasks being automated, and said the idea of agents built to manage users and schemes concerns him. His conclusion: “It’s not going to replace them, because first of all, AI makes mistakes.”

Our reading: all three describe the same job from different angles. The configuration part shrinks. The judgment part, saying no, keeping things simple, bringing people along, grows.

Data Center ends in 2029. What now?

Here the champions disagreed in a useful way.

Peter’s view is the bold one. He pointed out that Data Center and Cloud are very different platforms, so the decision belongs to the CIO or CTO, and non-Atlassian options exist too. On one point he was firm: you could hope Atlassian extends Data Center after 2029, “but hope is not what companies are building on.”

Deepanshu, who has helped hundreds of teams move, added the part people forget: “It’s not only about the technicality; it’s also about change management, it’s also about people management.”

Six questions we’re still thinking about

Keynotes give answers. The best part of an event is the questions you take home. These are ours.

1. If AI never says no, who in your organization does? Deepanshu’s custom-field example points to a real gap. Agents are built to be helpful, and helpful can mean 700 workflows. Someone has to own the word “no.”
2. What happens when 85% and 6% meet? Atlassian’s Leading with Context report says 85% of knowledge workers use AI, yet only 6% of executives can point to clear organization-wide ROI. Its engineering study found the same 6%. Even Atlassian admits usage counts can mislead, which is why it now measures AI superusers by function. Nikola Perišić put the human side simply: “Should I trust the AI itself or not?”
3. When an agent “runs as user,” who is accountable for its mistakes? AMP gives every agent an owner and an audit trail. Good. But if the owner was in a meeting when the agent changed 40 work items, is the owner the person who approved it, or the person who configured it?
4. Will AMP become an open standard, or stay an Atlassian protocol? MCP spread because many vendors adopted it. AMP’s value grows if agents from other platforms speak it too. Atlassian hasn’t named outside partners yet.
5. If DX can show a price per pull request, when does a human become the cheaper option? Measuring AI cost is overdue. It also invites a comparison many teams have avoided so far.
6. What does a 12-week engagement say about how hard this is? Atlassian’s new Forward Deployed Engineering program puts senior AI engineers inside customer teams for about 12 weeks. Expedia reports roughly 3,200 hours saved a year. Impressive, and also a signal that AI adoption at scale is a project, not a switch.

ikuTeam at Team ’26 Europe: three days at booth 407

Our crew of eight wore orange shirts that said what we do in six words: SharePoint files, connected. Jira workflows, automated.

The conversations followed the theme of the event. If AI needs context, and a big part of company context lives in SharePoint, Google Drive, and other cloud storage, how do you bring those files into Jira and Confluence without copying them, losing permissions, or exposing what shouldn’t be seen?

The question we heard most from attendees, customers, and partners was a fair one: “Atlassian already connects to SharePoint and Google Drive. Why would I need you?”

Our answer: Atlassian’s integration and ours do different jobs, and most teams use both. Atlassian’s Smart Links and Teamwork Graph connectors make documents discoverable, so Rovo can find a file wherever it lives. They do that well, and they respect Microsoft permissions. Our apps connect the right file to the right page or work item, and let teams act on it without leaving Jira or Confluence.

What we showed:

• SharePoint Connector for Jira: auto-create SharePoint folders from manual, scheduled, or status triggers, move attachments out of Jira storage, and link folders to the JSM customer portal.
• SharePoint Connector for Confluence: browse, upload, rename, and edit SharePoint files on the page, with changes saved back to SharePoint. Files from Microsoft Teams channels work too.
• ikuTeam Office for Jira and ikuTeam Office for Confluence: one-click editing of Word, Excel, and PowerPoint attachments, with real-time collaboration and no Microsoft license required. The Advanced Edition adds PDF editing and annotation.

Teams using ikuTeam report reclaiming 20% of the workweek with file automation, saving 70% on Jira Cloud attachment storage, and spending 38% less time searching for internal files.

There’s a cost angle too. Our Head of Product, Nelson, laid it out in Confluence vs SharePoint: stop paying twice to edit a spreadsheet. His test for every document is one question: “Who else needs this file, and for how long?” If someone only needs a Microsoft seat to edit files attached to Confluence pages, that overlap is worth an audit.

Where does the data live? Files stay in SharePoint, Google Drive, or wherever they already are. Nothing is copied into Jira or Confluence; permissions follow your storage by default, and our services run in the US and EU, routed by your data residency setting. Atlassian’s Third-Party Folders announcement shows the platform and the Marketplace pulling in the same direction: files belong where the work happens, with context that people and agents can both trust.

Thank you to Peter, Nikola, Deepanshu, and Anahit for sitting down with us, and to everyone who came by for a demo, a question, or some swag.

What’s next: Team ’27 in San Francisco and Copenhagen

The next Atlassian Team is Team ’27, April 20 to 22, 2027, at the Moscone Center in San Francisco. The European edition follows on October 5 to 7, 2027 in Copenhagen, and you can already sign up for the save-the-date to hear when registration opens.

If Amsterdam set the rules for agents on the team, San Francisco will show whether teams used them. We’ll be watching three things: whether EU AI inference has shipped, whether AMP has partners beyond Atlassian, and whether that 6% has moved.

We’ll leave the last word to the people. Peter Van de Voorde told us about a partner-led event on the Sunday before the conference, where more than 100 people worked together on the future of the ecosystem. “It’s that power, that feeling of community, that is still going to drive us into a people-first AI world,” he said. “If it’s an AI-first world, then we’re just going to be the handbrakes on evolution.”

Anahit Sukiasyan said it shorter: “You are coming for the tools, but staying for the people.”

See you in San Francisco.

FAQ: Atlassian Team ’26 Europe

When and where was Atlassian Team ’26 Europe?

Atlassian Team ’26 Europe took place on October 6 to 8, 2026 at Amsterdam RAI in the Netherlands.

What was the biggest announcement at Team ’26 Europe?

AMP, the Agentic Multiplayer Protocol. It defines how AI agents work alongside people, with an owner, a profile, scoped permissions and a full audit trail for every agent.

Can I watch the Team ’26 Europe keynotes on demand?

Yes. Keynotes and featured sessions are on Team ’26 Europe Digital with a free sign-up. The closing keynote is available until November 7, 2026.

Is Atlassian processing AI in the EU now?

Atlassian announced local EU AI inference, which restricts LLM processing to models hosted in the EU. No availability date was shared at the event.

When is Atlassian Team ’27?

Team ’27 runs April 20 to 22, 2027 at the Moscone Center in San Francisco. Team ’27 Europe is October 5 to 7, 2027 in Copenhagen.

When does Atlassian Data Center reach end of life?

On March 28, 2029, impacted Data Center products become read-only. Existing customers can’t buy new subscriptions or expansions after March 30, 2028. Bitbucket and Bamboo Data Center are not included.