The single slide that reframes the next 12 months of enterprise AI. Mike said it. Now everyone has to answer it.

Friends, I'm writing this from my seat at Team '26 in Anaheim. Front row, no notes app open, just trying to absorb what just happened on stage.

Mike Cannon-Brookes, Co-Founder & CEO, Atlassian opened the keynote with one slide. Acceleration equals Context times Intelligence. He paused, let it sit, then said the line that's going to get quoted for the rest of the year: in 2026, anyone can buy smarts by the token. What you can't buy is your team's institutional memory.

That's the whole bet. Let me walk you through what they shipped to back it up.

I made you a map. Save this image, share it with your team, send it to the CDO who couldn't make the trip. Seven announcements, four headline numbers, organized so you can see what actually matters in 30 seconds.

Everything else at Team '26 sits on top of this. So I want to spend a minute on it.

Atlassian's been quietly building something they call the Teamwork Graph for years. It's the connective tissue between your people, your work, and your tools. Every Jira ticket, every Confluence page, every Loom video, every Slack thread that connects through their integrations — it all becomes a node in a graph. Atlassian says it now holds more than 150 billion connections.

For most of its life, only Atlassian's own products could read it. Today that changed.

Three ways to plug in:

Teamwork Graph CLI gives developers terminal access. Pipe context into Claude Code, Cursor, or whatever coding agent your team uses. 300+ commands. Open beta starts now.

Rovo MCP Server exposes Teamwork Graph tools to any MCP-compliant agent. ChatGPT, Claude, anything that speaks the protocol can now reason over your Atlassian context. Open beta.

Teamwork Graph Connectors via Forge are now generally available. Build a connector to your proprietary system — your internal CRM, your custom pipeline tool, your industry-specific platform — and that data lights up across Rovo and Atlassian Analytics with permissions intact.

My take from the front row: this is the most consequential thing Atlassian announced this week. They're not trying to win every agent surface anymore. They're trying to be the context layer every other agent runs on. That's a much bigger market and a much harder thing to displace.

The headline number Atlassian put on the wall: agents grounded in the Teamwork Graph delivered 44% more accurate results using 48% fewer tokens in their internal benchmarks. I've been asking enterprise AI teams about exactly this gap all year. This is the cleanest answer I've seen anyone give.

The Mercedes-Benz case study is the proof point everyone's been waiting for. They built custom Forge connectors for defect management, requirements traceability, and release workflows. The connectors plug their automotive-specific systems into the Teamwork Graph. The result: 90% better defect intake quality, 85% faster duplicate detection, 10x faster software delivery.

Read that again. Ten times faster. That's not a pilot. That's production.

There's also a new explorer site at TeamworkGraph.com where you can visualize what your own graph already knows. I tried it backstage. The "people you collaborate with most" view alone is worth the click.

The numbers Atlassian put up for Rovo are now hard to dismiss. 75% of the Fortune 500 use it. More than 90% of their enterprise customers. 14 million Rovo-assisted actions just last month. Agentic automations grew 7x in the last six months across customers.

What got announced today:

Rovo Studio is generally available. The pitch is that anyone — not just engineers — can build agents, automations, and Forge apps from natural language. Roles, approvals, versioning, audit logs all built in. This is Atlassian's answer to shadow AI: give people a sanctioned, governed way to build, or watch them build it themselves outside your perimeter.

Max is a new reasoning mode coming to Rovo Chat. Hand it a messy ask, it builds a multi-step plan, runs it across your stack, drafts the doc, updates Jira items, finds time on calendars, and loops you back when it needs you. Coming soon in early access. This is the moment Rovo crosses from assistant to autonomous teammate.

Code Intelligence in Rovo is in early access. Engineers can ask intent-level questions across multi-repo environments. The example Atlassian gave on stage: "which services still use an outdated UI pattern and who owns the migration plan?" That's a question grep can't answer. That's a question a human teammate who's been on the team for two years can answer. Now Rovo can too.

Agents in Jira are GA. You can assign Jira work to Rovo or to third-party agents from Cursor, GitHub Copilot, Figma, Canva, Gamma, Amplitude, and others. Comment iteration, audit trails, admin controls. The change that matters most for software teams: agent work becomes visible in the same tracker as human work. Single-player AI just became multiplayer.

What I keep coming back to: if you talk to engineering leaders right now, every one of them is asking how to measure AI ROI inside the SDLC. DX — the company Atlassian acquired last year — just shipped AI Code Insights, Agent Experience, and AI Pulse. Where AI is generating code, how agents are performing, what the impact on productivity and reliability looks like. This is the dashboard CTOs have been asking me about all year.

This is the launch product leaders should pay closest attention to. I'd argue it's the most strategic announcement of the day if you're not in engineering.

The thesis is sharp: shipping used to be the hard part. Agile, DevOps, CI/CD made delivery cheap. Now AI is making building itself almost free. Prototypes that took weeks take hours. The bottleneck moved.

What's the new bottleneck? Deciding what to build.

The Product Collection bundles four things into one connected system:

Jira Product Discovery Enterprise is also coming, with the governance and compliance controls regulated industries need. Decisions traceable from initial signal through to shipped work via the Atlassian Data Lake. If you're in financial services, healthcare, or anything with serious audit requirements, this is the version to watch.

Currently in early access. If you run a product org, get on the waitlist this week.

The Teamwork Collection (Jira, Confluence, Loom, Rovo) got the biggest set of feature updates of any single Collection. Quick tour:

Remix with Rovo turns text on a Confluence page into charts, infographics, timelines, geo maps, and quadrants without leaving the page. Atlassian shared a stat I want to repeat: Confluence pages with visual elements are nearly twice as likely to get read by a wider audience. Now in beta.

Confluence Slides is in beta this month. AI-generated decks grounded in your Teamwork Graph. Ask Rovo to create or edit slides, it determines structure, writes content, builds the charts. Honestly, I want this for my own newsletter prep.

Loom now does agent briefings. Record a walkthrough of your requirements or feedback. What you say, show, and click gets captured as multimodal input and translated into a structured prompt that agents can act on. The output: a suggested action plan that turns into Jira work items in one click. We've been briefing teammates with Loom for years. Now we're briefing agents the same way.

Bug reporting with Loom and Jira is GA. Capture device info, console logs, network data automatically, then assign to Rovo Dev to draft the fix.

The shared thread across all of these: every announcement here works better because Jira, Confluence, and Loom share the same Teamwork Graph foundation. The graph is doing the heavy lifting. The features are just the surfaces.

Jira Service Management added three new AIOps partners: Lansweeper for cyber asset intelligence, Coralogix for full-stack observability, and Honeycomb for high-cardinality debugging. They join New Relic, Dynatrace, and BigPanda.

The pattern across all three integrations is the same. The Rovo Ops agent connects to the partner via MCP, pulls their telemetry directly into the incident, generates root cause hypotheses, and auto-builds post-incident reviews in Confluence.

For change requests, the Lansweeper integration generates approval recommendations with rationale and auto-populates the risk section. That's the kind of automation that saves IT ops teams hours per ticket.

The pattern I want you to notice: every one of these AIOps partners exposes their data via MCP. Atlassian doesn't have to build deep custom integrations anymore. They just speak the protocol. This is the year MCP went from "interesting standard" to "table stakes for enterprise software."

This is incident response moving from swivel-chair (alert in tool A → log in tool B → trace in tool C → ticket in tool D) to single-pane. The integrations are in early access.

Dia, the browser Atlassian acquired last year, is now ready for teams. SOC 2 Type II, single sign-on, Chromium MDM support, layered defenses against prompt injection. A closed beta is starting for advanced enterprise features including Guard integration.

The feature to notice is Morning Brief. While you sleep, Dia pulls from your overnight Slack, your calendar, and your action items, and assembles a brief for you. Most of your day already happens in browser tabs. Dia turns the browser itself into the agent surface.

I've watched five different companies try to make the AI browser happen this year. Dia inside Atlassian's enterprise distribution might actually be the one that lands.

I want to give you three takeaways before I run to the next session.

One. Context is now the official Atlassian thesis. Not a feature, not a marketing line, the entire architectural bet. They're betting that owning the graph of how your company actually works is more durable than any model choice. Hard to argue with the math when grounding agents cuts tokens almost in half and lifts accuracy 44%.

Two. Opening the Teamwork Graph via MCP and CLI is the most consequential decision Atlassian made this week. It changes the competitive frame. They're no longer fighting Microsoft, Google, and OpenAI for the agent surface. They're trying to be the context layer that every one of those agents runs on. Different game. Bigger market. Harder for any one model vendor to disrupt.

Three. The Mercedes-Benz numbers are the proof point this whole industry has been waiting for. Custom connectors plus a context graph plus agents equals 10x software delivery. Not a benchmark. Not a demo. Production at one of the most demanding engineering organizations in the world.

If you're a CDO, CTO, or product leader watching this from outside the conference, here's the question to take into your next leadership meeting: where does your institutional context live today, and would your agents be smarter if it lived somewhere they could reach?

If the honest answer is "it's scattered across 40 SaaS tools and nobody owns it," then you've found your 2026 priority.

I'll be on the floor at Team '26 all week, doing interviews with the product leaders behind these announcements. Reply and tell me which one you want me to dig into first. Best replies get featured in the next issue.

If this was useful, forward it to one person on your team. The newsletter grows because of you, and I keep getting front-row seats because of the audience you've built with me.

— Ravit Jain
Founder & Host of The Ravit Show
Live from Anaheim · May 6, 2026