A deep dive into the Team '26 keynote, where Atlassian unveiled how Context - not Intelligence - will define the winners of the AI era.
When Atlassian CEO and co-founder Mike Cannon-Brookes opened Team '26 in Anaheim, he didn't talk about new features first. He talked about a fundamental shift: the birth of what he called "a new species" - the AI Native Organization.
This isn't a traditional business with a few chatbots floating around. It's a completely new way for organizations to operate and deliver results. Humans focus on what matters most - setting direction, making judgment calls, and resolving ambiguity - while AI agents handle routine execution at scale.. Tools and tasks change, but the purpose remains the same: solving problems that matter.
Cannon-Brookes was direct about the urgency. While model performance is improving exponentially, many businesses are still stuck in linear gear, piloting endlessly or waiting to see how things play out. He didn't mince words: "That's not caution. That is surrender in slow motion. You cannot wait to see your way through an existential shift in technology."
Explore more: Transforming Organizations into AI-Native Enterprises
The most important insight of the keynote was a simple formula:
Acceleration = Context × Intelligence
In 2026, raw intelligence has become a commodity. You can literally buy smarts by the token. Models, therefore, cannot be your competitive edge. The differentiator is Context - the institutional memory of every project, every goal, and every workflow your team has completed.
Intelligence is the engine, but Context is the fuel. As Cannon-Brookes put it: "If your AI doesn't know what choice you made in 2024, it can't help you win in 2026."
This reframing has profound implications. The companies that will win aren't the ones with access to the smartest models - those are available to everyone. The winners are the ones who have systematically captured, connected, and made accessible the collective memory of their organization.
Atlassian's answer to this challenge rests on four interconnected layers:
The System of Work is the connective foundation - how teams align to goals, plan their work, and unlock the organization. It's the operating model that turns the formula into practice.
The Teamwork Graph is the data fabric beneath everything - the context layer. Think of it as a living map - every Jira ticket you close, every Confluence page you edit, every Slack thread you participate in becomes a connected piece of your organization's memory.
The AI Gateway is the secure portal connecting your private organizational context to leading-edge AI models, without compromising data sovereignty.
Rovo is the interface that turns context and intelligence into momentum - bringing context-rich AI to every individual: the builders, the knowledge workers, the leaders.
To demonstrate how the Teamwork Graph operates, Chief Product and AI Officer Tamar Yehoshua joined Cannon-Brookes on stage. The demo started with Rovo's upgraded Memory capabilities, which now include two distinct types: implicit memory that continuously learns from the Teamwork Graph, and explicit memory that users can manage directly.
Cannon-Brookes asked Rovo to prepare for a meeting with ServiceRocket - an Atlassian partner of more than two decades. In about three minutes, Rovo pulled from 61 different sources across the graph, generating a real-time BI dashboard of the 20-plus year customer relationship, complete with charts, stakeholder maps, and use case recommendations.
The scale is staggering. Atlassian is now ingesting billions of objects per week into the Teamwork Graph. Image content and text inside images are searchable. Microsoft Teams meeting transcripts are indexed alongside messages. Salesforce campaigns, cases, and contacts come in with field-level controls. The synchronization target is under 10 minutes for any change. And customers are already running 5 million agent invocations per month - and growing rapidly.
But the Teamwork Graph isn't just for software teams.
One of the most compelling examples came from Atlassian Williams Racing. Roughly 50% of Atlassian customers deal with physical objects as part of their core business - logistics, manufacturing, vehicles, satellites. Williams has modeled every component of a Formula 1 car as a first-class object in the Assets app.
Those parts are no longer trapped in spreadsheets. They're connected to work, people, code, service requests, projects, and 48 years of institutional knowledge. Between races, when a race engineer considers upgrading a suspension bracket, they can ask Rovo whether it's worth doing and whether it can be completed in time. Rovo pulls from service requests, projects, and knowledge bases to give a recommendation grounded in real trade-offs.
As Cannon-Brookes noted: "No single person at Atlassian Williams actually had that context, but the Teamwork Graph did."
Head of AI and Product Craft Sharif Mansour unveiled one of the keynote's biggest announcements: a new Code Search Connector for Rovo, now in early access. It goes beyond simple keyword matching - it understands the purpose and relationships within source code.
The numbers behind Atlassian's internal usage are remarkable:
More than 11 million code files indexed
Approximately 1.5 billion lines of code
Spanning 25+ years of development history
Searched across Bitbucket Datacenter, Bitbucket Cloud, and GitHub
In one demo, Sharif asked Rovo to find every button in the Confluence codebase not using the latest design system, identify the people working on it, and surface the relevant style guides. Rovo identified the legacy and new libraries, produced an audit with adoption percentages, and listed the teams and Slack channels needed to coordinate the migration. A task that would have taken senior engineers days was completed in minutes.
In a more humorous demo, Sharif asked which TODO comments from Mike Cannon-Brookes himself were still in the codebase. Rovo found two "fossil TODOs" over 21 years old - one supporting Confluence 2.8, a version that predates even Datacenter.
The serious lesson behind the joke: for customers running Datacenter today, all of that institutional knowledge - every commit, every decision, every comment - comes with you when you migrate to Cloud and gets folded into the Teamwork Graph.
Sharif framed a critical shift in how software gets built: business processes were historically designed around the assumption that code takes time to produce. If code is no longer the bottleneck, then planning, judgment, and building the right thing the right way becomes the new constraint.
A live demo showed Rovo creating an AI Plan directly inside a Jira project for a financial services dashboard feature. Rovo invoked the Code Intelligence skill automatically, discovered the change spanned six repositories, and - when faced with a 50/50 architectural decision - paused to ask the user rather than guess. The plan was then broken down using the Intelligent Work Breakdown skill into tasks for human team members, Claude Code, and a custom agent. Claude Code automatically picked up its assigned tasks through Jira automation.
This is what Atlassian means when it calls Jira "the AI control plane" - a place where humans and agents work side by side, where automation rules orchestrate complex multi-agent workflows, and where the system understands what's happening across both worlds.
The major announcement: Agents in Jira is now Generally Available (announced at Team '26, April 2026). Any agent can connect via MCP. The GitHub Copilot agent ships out of the box, with Claude Code, Cursor, and OpenAI's Codex integrations coming in the following weeks.
Agents in Jira aren't just for technical teams. A pre-recorded demo showed a sales enablement team orchestrating multiple agents to handle their Revenue Kickoff prep - a Canva agent for creative assets, a Gamma agent for presentation generation, and a custom brand guidelines agent.
Their workflow was elegantly simple: drag a Jira card into the agent's status column, and the agent picks up the work. Each agent assignment creates a Rovo chat session that can be resumed anywhere - desktop, mobile app, on a bus, in a taxi.
For product managers facing a flood of new feature throughput from accelerated engineering teams, Atlassian announced a new Feedback App joining Jira Product Discovery in the Product Collection. Feedback flows in automatically from connected sources - Salesforce, Zendesk, Microsoft Teams, Pendo, or any MCP - and Rovo can generate ideas directly into the discovery backlog, automatically associated with goals.
For IT Ops and SRE teams, a new Incident Command Center brings together GitHub change history, Datadog and New Relic observability data, service dependency visualization, and automated mitigation steps. Forrester Consulting research found teams adopting the service collection save over 55 minutes per incident on average.
Perhaps the most strategically significant announcement was that the Teamwork Graph is now reaching beyond Atlassian's walls through two channels:
The Atlassian MCP Server - already one of the most-used MCP servers globally - now exposes the full Teamwork Graph. In a demo, Sharif played the role of a designer in Figma, asking Rovo to design an analytics dashboard for a retail company using only a Jira work item as input. Rovo traversed the graph, pulled related work items, Confluence specifications, and Google Docs, and generated a prototype using the correct design tokens - work that would have taken half a day of Slack messages and tab-switching.
The Teamwork Graph CLI is built primarily for AI agents and exposes more than 380 tools and commands. In one of the most ambitious demos, Sharif assigned Claude Code an unhelpful Jira ticket - "Mobile-63" with no links and minimal context - and asked it to research and visualize related context. Claude Code used the CLI to surface related work items not explicitly linked, previous pull requests, GitHub branches, Figma designs, Confluence docs, the relevant team members, and even an instance where Claude Code itself had previously contributed to a related ticket.
The benchmark results are concrete:
44% improvement in answer time in Claude Code when connected to the Teamwork Graph CLI
48% reduction in tokens used - direct cost savings
As Sharif framed it: "The organizations that will succeed won't be the ones who lock their context away. It'll be the ones who make sure that context is most reachable, most connected, most useful in all the apps they use."
The keynote closed with an unexpected partnership: Josh Miller, CEO of The Browser Company of New York, joined Cannon-Brookes to introduce Dia, a browser paired with the Teamwork Graph.
Dia's flagship feature, the Morning Brief, captures the design philosophy: you shouldn't need to prompt AI - AI should proactively help you. Each morning, Dia presents a personalized memo pulled from calendar, open tabs, browsing history, Slack, Teams, and the Teamwork Graph. It runs overnight while users sleep. There's no setup required because the browser is already logged into everything.
Dia can also generate interactive web pages on the fly. Josh demonstrated by asking Dia to build a personal Team '26 schedule combined with a Disneyland visit plan - the resulting tab knew his keynote slot, his coffee meeting with his mom afterward, and that Space Mountain is better at night. The point isn't the theme park, it's that the browser eliminates the gap between information scattered across tools and the decisions you need to make.
The strategic narrative running through Team '26 is clear and consistent: in an era when intelligence is commoditized, the moat is your context. Every connector that grows the Teamwork Graph, every agent invocation that adds back to it, every workflow that flows through Atlassian apps compounds the value of your organization's collective memory.
For organizations evaluating their AI strategy, the keynote framed the choice starkly. The companies that succeed in the AI Native era won't be those with the best models - those are available to anyone. They'll be the ones who treat their organizational context as a strategic asset, who connect their work systems rather than fragmenting them, and who orchestrate humans and agents together as integrated teammates.
As Cannon-Brookes summarized near the end of the keynote: "We are building the memory of the company that you are becoming."
The AI Native Organization isn't coming. According to Atlassian, it's already here - and the only question is whether your organization is building the context graph that will make it thrive.
Learn more from Atlassian’s sessions at Team ’26 California HERE.
