Stay updated with Data & AI News!
Join The Ravit Show Newsletter
BREAKING: Atlassian launches AMP and changes how teams work with AI
AMP, a rebuilt MCP server, an OpenAI partnership, Forward Deployed Engineers and 120+ launches. Here is what it means for your team
Ravit Jain
October 07, 2026
Breaking news from Amsterdam
Hey friends,
I am writing this from the floor of the RAI in Amsterdam, a few hours after the Founder Keynote at Atlassian Team '26 Europe.
Atlassian just announced AMP, the Agentic Multiplayer Protocol. It is their answer to a question every leader I meet is asking right now. How do people and AI agents actually work together as one team?
I have been covering enterprise AI for years. This is one of the clearest stories I have heard from any vendor this year. So let me walk you through all of it: what was announced, why it matters, and my honest take.
The big idea: AI is going multiplayer
Mike Cannon-Brookes opened the keynote with a simple observation, which he also wrote up in his founder update.
Most AI adoption today starts in single-player mode. One person typing into a chat window on their laptop. That person gets faster. The team does not.
Anyone who has led a team knows why. Work is a relay. Ideas get better when someone else catches what you missed. If your AI lives in a private browser tab, none of that happens.
The numbers back him up. Atlassian's new report, Leading with Context, found that 85% of knowledge workers use AI, while only 6% of executives can point to clear, organization-wide ROI. Their State of Teams survey of 12,000+ knowledge workers tells the same story. Individuals are faster. Business results are not catching up.
AMP is built to close that gap.
What is AMP?
AMP is the foundation for how humans and agents collaborate across the Atlassian platform. According to the official press release, it sets the rules of engagement so agents have the context, identity and governance to work safely as part of a team.
It has 3 layers.
1. In-the-flow collaboration. Agents show up where work already happens. You can @mention an agent in a Confluence page, debate an approach in a Jira comment thread, or brief one with a quick Loom video. Through MCP, tools like Figma and your IDE plug in too.
2. Identity and shared presence. Every agent has a clear owner and a distinct profile. When you open a doc or a whiteboard, you see the agent's cursor right next to your teammates. Developers running agents locally on their laptops get those sessions bridged straight into the Jira board.
My favourite demo of this was the Confluence version history. Every edit is attributed: Rovo with one person, Claude with another, Canva and Figma with someone else. You can see exactly which agent changed what, and who was behind it.
3. Context and governance. Every agent is grounded in the Teamwork Graph. Permissions are scoped: an agent can run as the user who prompted it, or on its own service account. Every action is logged, with human checkpoints built in.
AMP begins rolling out today. More at atlassian.com/amp.
The numbers that stood out to me
10M+ times a month, people and agents already work together on Atlassian
250B+ connected objects in the Teamwork Graph, across 7 context types
20+ new connectors announced on stage, in a library of 80+
~2M monthly active users on Atlassian MCP, with 15M+ tool calls a day
15x MCP growth in 6 months
Up to 25% fewer tokens on the new MCP server for the same Jira and Confluence work (internal benchmarks on Claude models)
44% higher quality answers using 48% fewer tokens when agents are grounded in the Teamwork Graph (Atlassian testing)
120+ enterprise capabilities shipped in the last year
Announcement 1: A rebuilt Atlassian MCP server
Atlassian rebuilt its MCP server from the ground up. It now covers 220+ tools across Jira, Confluence, Loom, Bitbucket, Goals, Code Search and soon Jira Service Management.
The part I like most: it does the whole job, beyond quick lookups. Ask it to catch you up on a ticket and it reads the full comment thread, the changelog and the linked pull requests. Ask it to turn a meeting Loom into a launch checklist and it drafts the Jira items and the Confluence page. If a teammate edits the same page while you work, it pauses so nothing gets overwritten.
Security is solid. OAuth 2.1 with PKCE by default, every request respects the user's existing permissions, and read, write and destructive actions are separated into risk tiers.
It is generally available now.
Announcement 2: A richer Teamwork Graph
Context was the word of the day. It was literally on the screen behind Mike and Tamar for a big part of the keynote.
The Teamwork Graph is Atlassian's institutional memory. Every closed ticket, decision and doc feeds it. This week it got big upgrades, detailed in Jamil Valliani's platform blog.
Rovo Code Search. The graph now reads source code down to functions, symbols and classes. Engineers can search across Bitbucket and GitHub in plain English without cloning a repo. Coding agents use the same context and burn fewer tokens doing it.
The best proof came from the Atlassian Williams F1 Team. On 1 real bug fix, Code Context cut their Claude Code run from $10.69 to $6.99, and from 26 minutes to 15 minutes. That is about 35% cheaper and 42% faster on a single task. Multiply that across an engineering org.
Atlassian Insights. Structured data is now a native context type. You can query your data lakes and warehouses in plain language, with catalog, glossary and lineage across 20+ structured sources. As someone who comes from the data world, this one made me smile.
Connectors. 20+ new connectors were announced on stage, including Zoom, Gong, Microsoft Entra ID and Google Identity, plus 65+ improvements to existing ones.
Artifacts. AI outputs from any tool, on or off Atlassian, get a permanent URL, enterprise permissions and get indexed back into the graph. No more good work dying in a chat window.
Announcement 3: Seeing what agents are doing
This is the part every CIO in my network will care about.
Agent Sessions in Jira. Local and cloud agent runs link automatically to the right work items. No more agents working in a terminal nobody can see.
Non-human identities. Every agent, app and service account, from Atlassian or anyone else, gets its own identity. Admins see what each one can access and who owns it, and can revoke access with one switch.
Agentic updates in Focus. Leaders get real-time summaries of cross-project progress, risks and blockers without chasing status reports.
Trust and compliance. Local EU AI inference keeps LLM processing on EU-hosted models. Rovo is aligned to ISO 42001 and the EU AI Act. Atlassian Guard Premium adds data scanning and guardrails across Rovo Chat and connected tools.
Announcement 4: New ways to work with agents
Rovo Work. A new mode in Rovo Chat for big, multi-step jobs. You give it a goal, it proposes a plan, you shape it, and it executes across Jira, Confluence and your connected tools. It can run for hours in a secure sandbox your admins control.
Record for Agent. Record your screen and voice in Loom, and it becomes a structured brief an agent can turn into Jira work items, a prototype or a Confluence spec. Now in open beta.
Interactive PR Reviews. Agents record a Loom walkthrough of their own code changes in Bitbucket, explaining what they did and the tradeoffs.
Third-party agent triggers. A status change in Salesforce can kick off an agent workflow in Jira or Confluence, with human checkpoints where you want them.
More in the Teamwork Collection update from Sanchan Saxena.
Announcement 5: Atlassian and OpenAI
Atlassian also announced a strategic partnership with OpenAI.
OpenAI frontier models power reasoning across Rovo. Teams can connect ChatGPT and Codex to the Teamwork Graph through the Atlassian plugin and MCP. DX gives engineering leaders visibility into how tools like Codex affect delivery speed and team health.
Inside Atlassian, 3,000+ developers already use Codex through ChatGPT Enterprise, connected to Atlassian MCP.
Put that next to the Claude Code result from Williams and how widely Atlassian MCP is used on Claude, and the message is clear. Atlassian wants to be the context layer for every model and every agent, whoever builds it.
Announcement 6: Forward Deployed Engineers
This one is easy to miss, and I think it matters a lot.
Atlassian launched a Forward Deployed Engineering program. Senior applied AI engineers embed directly with customers to build production AI: enriching the context in the Teamwork Graph, building custom Rovo agents and automations, and fixing workflows across technical and business teams.
Avani Prabhakar, Chief People and AI Enablement Officer, put it well. The value in AI has shifted from access to models to whether AI understands your business context. Accounts working with Forward Deployed Engineers are seeing 50% higher AI adoption.
How it works:
Typical engagements run 12 weeks
Customers put their own engineers alongside Atlassian's, so the skills stay in-house after the engagement ends
Work covers custom agents for IT, HR and reporting, search and connectors, Jira automation like bug triage and backlog cleanup, Confluence content operations, and enterprise rollout
Fixes, reusable agents and connectors built for 1 customer flow back into the platform for everyone
The customer stories are real. Expedia archived 600,000 Confluence pages, saves about 3,200 hours a year, and now runs 10,000+ agent invocations a month across 50 live projects. Reddit is working with the team on its enterprise AI transformation using Rovo and the Teamwork Graph.
Atlassian also released the Forward Deployed Engineering Playbook, with lessons from the program for anyone thinking about embedding FDEs.
My view: tools alone never close the gap between 85% usage and 6% ROI. People who sit inside your workflows do. This is the human half of the AMP story.
Conversations from the floor
I got time with Tamar Yehoshua, Atlassian's Chief Product and AI Officer. What stayed with me is how much thought went into the design. Atlassian is treating agents as teammates, so the experience has to work for mixed teams of people and agents. Small details like seeing an agent's cursor on a page sound simple. They change how much you trust what is happening.
Sherif Mansour, Head of AI, brought the product to life on stage alongside Mike and Tamar. Watching a short Loom turn into real Jira work items made the whole story click for me.
Full interview coming soon on The Ravit Show.
My take: why this is a game changer
I have done 750+ interviews with founders, CIOs, CTOs and CDOs. Here is the pattern I keep seeing.
The model is rarely the blocker anymore. Teams can rent great intelligence from OpenAI, Anthropic or Google. What they cannot rent is the context of how their business works, and the trust to let an agent act on it.
That is exactly where AMP lands. 4 reasons I think it matters:
1. It makes agents accountable. Owner, identity, permissions, audit trail. That is the checklist every security team asks for, and now it ships by default.
2. It makes agent work visible. Agent sessions in Jira, cursors on the page, attributed version history, Loom walkthroughs of code. Leaders can finally see what is happening.
3. Context pays for itself. The Williams F1 result and the 44% quality jump with 48% fewer tokens show that better context means cheaper, faster, better agents.
4. It is open, and it comes with people. MCP, Claude, ChatGPT, Codex, Cursor, third-party agents, plus Forward Deployed Engineers to help you actually get it into production.
What I will be watching: how fast enterprises move from pilots to agents with real authority, how pricing for agent work evolves, and whether AMP becomes something other vendors adopt.
Read more
One question for you before you go. If an agent joined your team tomorrow, would you know what it was doing?
Hit reply and tell me. I read every one.
See you next week,
Ravit Jain
Founder and Host, The Ravit Show
