I came away from Atlassian Team ’26 Europe in Amsterdam with plenty to think about.
There were some genuinely interesting product announcements around Rovo, AI agents and Atlassian’s wider platform. Above all, though, the event made my colleague Martin and me think more deeply about something much more foundational: how we manage work, and what becomes possible when the context around that work is connected.
One question in particular has stayed with me:
What becomes possible when AI understands how an organisation works, rather than only the prompt or document immediately in front of it?
It was also brilliant to spend time with the Atlassian Foundation team and this year’s non-profit cohort. The conversations between sessions were every bit as valuable as the sessions themselves: comparing experiences with organisations working in very different contexts, finding common challenges, and learning from how others are using technology to support their missions.
The conversations reinforced something important. Non-profits may share many technology challenges with commercial organisations, but our operating context is often different.
We work within constrained budgets, across distributed teams, volunteers, partners and communities. Some organisations have established technology functions; others depend on a small number of people carrying several roles. All of us need to make careful choices about where technology can genuinely increase impact and where it could introduce unnecessary cost, complexity or risk.
Against that backdrop, these are the questions I brought home from Amsterdam.
1. What changes when AI becomes part of the team?
Most of our AI use today is still essentially single-player.
One person opens a chat window, explains what they are doing and works privately with an assistant. The final output might eventually reach Jira, Confluence or another shared system, but the interaction that shaped it remains largely invisible to the wider team.
Atlassian describes the alternative as multiplayer AI.
The idea is that assistants and agents increasingly participate where teamwork already happens. Their contributions are visible. They can take on defined tasks, draw on the surrounding context, and operate within permissions and boundaries set by our organisation.
This is wider than agentic AI alone:
- Agentic AI is about what AI can do: plan, act and work towards an outcome.
- Multiplayer AI is about how people and AI work together: shared context, visible contributions, hand-offs and accountability.
- AMP, the Agentic Multiplayer Protocol, is Atlassian’s framework for bringing those ideas together inside team workflows.
For organisations like ours, I think that distinction matters.
The opportunity is not simply to automate more activity. It is to improve how our work is coordinated, how knowledge is retained and how colleagues can build on work previously undertaken by both people and AI.
If someone spends an hour investigating a difficult problem with an AI assistant, how much of that work becomes organisational knowledge?
Too often, only the final answer survives. The reasoning, sources, rejected options and surrounding context disappear when the session closes.
A multiplayer model suggests something different: work undertaken with AI becoming part of the visible memory of our teams.
That could be particularly valuable for us at Mary’s Meals, where capacity is limited and the loss of organisational knowledge can have a disproportionate effect.
2. Is context becoming the more important AI advantage?
One of the central ideas from the keynote was expressed through a simple formula:
Acceleration = Context × Intelligence
As access to capable AI models becomes more widespread, the model itself may become less of a differentiator.
The harder part is creating, connecting and making available meaningful organisational context:
- Who owns this area of work?
- What outcome are we trying to achieve?
- What decisions have already been made?
- Which work depends on this?
- Where is the current and authoritative knowledge?
- Who else is working on something related?
- What is the AI permitted to access or change?
Atlassian sees its Teamwork Graph as the layer connecting this context across work, knowledge, people, goals, code and services. Atlassian says the graph now connects more than 250 billion objects and relationships.
That changes how I think about AI readiness.
AI readiness is not only about giving people access to AI tools or helping them use those tools effectively. It also depends on whether our organisation has preserved enough trustworthy context for AI to work from.
For many of us, that context may currently be spread across:
- task boards and project-management tools;
- spreadsheets maintained by individuals;
- documents and shared drives;
- email and chat conversations;
- specialist fundraising, programme or case-management systems;
- meeting notes; and
- knowledge held primarily in people’s heads.
That fragmentation already makes work harder for people. AI may simply make its consequences more visible.
An AI system cannot reliably connect a decision, its owner, the work that followed and the knowledge created if those things have not been recorded or connected in the first place.
Simply put:
AI makes good information and work management more valuable, not less.
3. Can a graph understand work it cannot see?
This was perhaps the most unexpected takeaway for Martin and me.
It probably should not have been unexpected. Good work management has always mattered. But Team ’26 concentrated our thinking on just how foundational that capability could become as AI draws on more organisational context.
A context graph can only understand the relationships represented within it.
If projects, tasks, decisions, dependencies and ownership are spread across numerous disconnected tools, then the resulting organisational context will also be fragmented.
That introduces another dimension to the case for more consistent work management.
Historically, an organisation might standardise on Jira to improve reporting, reduce tool sprawl or make project delivery more consistent.
In an AI-enabled organisation, there may be another benefit: better, reusable context.
The effect could become cumulative:
More work represented consistently → richer relationships between that work → better context for people and AI → more useful assistance and automation.
I don’t think this has to mean putting everything into Jira.
We will continue to need specialist systems of record. A fundraising platform, finance system, HR platform or programme-management application may remain the authoritative home for the data and processes it supports.
What crystallised for me in Amsterdam was the distinction between a system of record and a system of work.
A specialist application may hold authoritative information about a supporter, grant, employee, programme activity or financial transaction. A system of work can describe the activity around it:
- What needs to happen?
- Who owns it?
- What does it depend on?
- What decision has been made?
- Where does it currently stand?
- Which organisational goal does it support?
For us at Mary’s Meals, this raised a genuine strategic question:
Should Jira increasingly become our preferred platform for structured work management beyond the teams already using it?
We have not answered that question, and I would not advocate an organisation-wide rollout simply because Atlassian has presented an attractive AI vision.
Wider adoption would require real effort. It would mean helping teams understand the value, designing proportionate ways of working, supporting colleagues through change and avoiding unnecessary complexity. We would also need to consider licensing and how Jira fits alongside tools people already use.
But Team ’26 made me think that we should be more deliberate about how and where we introduce new work-management tools.
When a team needs structured task or project management, perhaps the question should not only be, “Which tool best fits this team?”
It should also be:
Will this choice contribute to, or further fragment, our shared organisational context?
4. What might Atlassian’s wider platform direction enable?
Atlassian’s move towards Collections made more sense to me when I viewed it through this lens.
Collections bring applications and AI capabilities together around types of teamwork and outcomes, rather than presenting each application as a disconnected product.
Stepping back from the individual announcements, five parts of the wider picture started to fit together for me:
- applications where teams undertake work;
- a graph that understands how that work relates;
- AI grounded in that context;
- protocols through which Atlassian and third-party agents can participate; and
- identity, permissions and governance around their activity.
Not every element is equally mature, and Atlassian is not the only enterprise provider developing a context layer.
Most of us also use several platforms. Microsoft 365, Salesforce, specialist programme systems and other applications will not disappear simply because an organisation adopts Jira and Confluence.
We may therefore never have one universal organisational graph.
The practical question for me becomes:
How do we connect the right context across platforms while preserving permissions, provenance and authoritative sources?
I find that more useful than debating which vendor has the “best” graph.
5. Where might non-profits find practical value?
The most credible opportunities will vary considerably between organisations.
I’m increasingly wary of beginning with, “Where can we deploy an agent?”
A more useful starting point might be:
Where do people currently lose time because knowledge, ownership or related work is difficult to find?
Several patterns seem worth exploring.
Finding trusted organisational knowledge
An internal assistant grounded in approved policies, procedures and guidance could help colleagues find authoritative information without relying on personal knowledge of where it is stored.
The difficult part is not necessarily creating the assistant. It is ensuring that the underlying information is current, owned and appropriate for the intended audience.
Improving hand-offs
Many non-profit processes cross organisational boundaries.
A request may begin with a supporter, volunteer, partner or programme team and require contributions from several internal functions. Connecting the original request to the work needed to resolve it could reduce manual hand-offs and make ownership clearer.
Preserving project history
Non-profits often rely on fixed-term funding, volunteers, partners or colleagues carrying several responsibilities.
Preserving decisions, dependencies and learning alongside project work may reduce the disruption caused when people change roles or responsibilities.
Making cross-team dependencies visible
A programme, campaign, organisational change or strategic priority may depend on several teams.
Representing that work consistently can help people identify overlaps and dependencies before AI is introduced. AI may then make those relationships easier to discover and use.
Starting with focused AI use cases
Rovo Studio and similar tools create opportunities to automate repetitive work or build internal agents.
My instinct is to start with clearly defined problems, trusted information and a human owner, rather than broad autonomy.
6. Connected context also creates connected risk
There is an obvious tension in all of this.
The richer the context available to AI, the more useful it may become. But the greater the context and agency, the more important identity, permissions, data protection and human accountability become.
If an AI agent can read, create or update organisational information, I think we need to treat that as a real organisational integration, not simply as another convenient AI feature.
For non-profits, this may involve particularly sensitive information about supporters, volunteers, employees, partners or the communities we serve.
More context should not mean universal access.
Agents that can take actions should have:
- a defined and documented purpose;
- a named human owner;
- access limited to what the role requires;
- clear boundaries around permitted actions;
- appropriate audit information;
- a route for human review or escalation; and
- a regular review of whether the agent remains useful and appropriate.
The mechanism can be proportionate to the organisation. Some may have a formal AI governance framework; others may use a simpler register and approval process.
The important point is that accountability remains clear.
7. What could non-profits do now?
None of this leads me to think that non-profits need to race towards autonomous agents.
Several useful foundations require little or no advanced AI.
Map where structured work happens
Identify the tools teams use to manage projects, tasks, decisions and dependencies.
The aim need not be to eliminate every variation. It is to understand where fragmentation creates a genuine barrier to shared context.
Make ownership clear
Important projects, services, processes and knowledge should have identifiable owners.
A graph may be able to connect information, but it cannot create meaningful accountability where none exists.
Strengthen the underlying knowledge
Review whether key knowledge is current, findable and clearly authoritative.
AI cannot reliably compensate for contradictory, outdated or ownerless information.
Choose bounded experiments
Start with a real problem, trusted source information and a clear measure of usefulness.
“Deploy an AI agent” is not an outcome. Reducing the time colleagues spend locating approved guidance might be.
Treat AI as an operating-model question
Teams need to understand how AI contributions will be reviewed, handed over and incorporated into normal work.
The technology becomes much more interesting when those expectations are shared by the team rather than decided independently by each user.
Questions for the community
Amsterdam left me with more questions than firm conclusions, which I think is healthy.
I would be interested in how other non-profit and social-impact organisations are approaching some of these areas:
- Where does structured work live in your organisation today? Is there a common work-management platform, or is it intentionally distributed?
- How broadly do you use Jira today? Is it concentrated in particular teams, or are you already using it as a wider organisational system of work?
- How do you balance the flexibility individual teams need with the value of shared organisational context?
- Which non-profit use cases genuinely benefit from AI agents, rather than conventional automation or better knowledge management?
My takeaway
The strongest message I took from Team ’26 Europe was not simply that Atlassian had added more AI to Jira and Confluence.
It was how the different parts of the vision fitted together.
People and AI working alongside one another. The Teamwork Graph contributing organisational context. AMP providing a framework for bringing agents into the team. Collections drawing applications and capabilities together around what teams are trying to achieve.
There is still plenty to mature, and we will each need to judge where the value justifies the cost and effort involved.
For me, the most practical takeaway is also the simplest:
Manage work well. Preserve the context around it. Make ownership clear. Keep knowledge trustworthy.
If the future really is teams of people and AI working together, those foundations will serve both.
All of this should ultimately be judged by something more important than technical sophistication:
Does it help our people collaborate more effectively and direct more of their time, energy and resources towards achieving the mission we are here to serve?
I’d be particularly interested to hear how other non-profits are approaching the system-of-work question. Are you trying to connect more of your organisation’s work in Jira, deliberately keeping work distributed across different platforms, or still working through what makes sense for your organisation?
