Skip to content
Governance

Atlassian now gives every AI agent an identity. Start with the agent inventory.

At Team '26 Europe in Amsterdam, Atlassian announced Agentic Multiplayer Protocol (AMP): agents take part in work with an identity an admin assigns, authority you scope, shared context, and results you can review. Agent accounts and a non-human identity inventory, with a single switch to revoke access, are coming soon. MSPs should start with a per-client agent inventory, then an access review.

Dark-ink Consultiply resource guide cover titled Atlassian now gives every AI agent an identity. Start with the agent inventory, with Atlassian mark, AMP identity/scope/review flow, coming-soon inventory and revoke switch callouts, and MSP inventory-first strip.

Published October 7, 2026 · ~7 min read

What Atlassian announced

At Team '26 Europe in Amsterdam, Atlassian announced the Agentic Multiplayer Protocol (AMP). Per Atlassian's blog (Jamil Valliani, Head of Product, AI), AMP "defines how agents take part in multiplayer work with an identity, scoped authority, shared context and tasks, and results you can review." Atlassian says AMP gives every agent "the context to be useful and the limits to stay safe, with an identity you assign and authority you scope." The Business Wire release dated Oct 7, 2026 frames AMP as rules of engagement so agents have context, identity, and governance to work as part of a team.

Coming soon, Atlassian says, are agent accounts and a non-human identity inventory. Per the blog, that gives every agent, app, and service account a non-human identity so admins can manage autonomous actors from one place, "including revoking access with a simple switch." That inventory and revoke switch are not live yet. Label them as Atlassian's roadmap, not a control you can click today.

Atlassian also says humans and agents already collaborate over ten million times a month on the platform. The release states Atlassian MCP has nearly 2 million monthly active users and handles over 15 million daily tool calls, a 15x jump in six months, with up to 25% fewer tokens for the same Jira and Confluence work in internal benchmarking on Claude models. Other product notes from Atlassian include Rovo Work (goal, plan, human shape, then execute across Jira, Confluence, and connected tools in an admin-governed sandbox), agent sessions in Jira that link local and cloud runs back to work items, and an Atlassian MCP rebuild from dozens of tools to 200+.

diginomica (Phil Wainewright, Oct 7, 2026) reports Mike Cannon-Brookes describing AMP as "a diplomatic protocol" more than a technical one, building on A2A and MCP. diginomica also relays four modalities Atlassian wants clear in the UI: human activity; agent with a non-human identity (agent-dedicated account); agent on behalf of a human; and human assisted by an agent. diginomica discloses that Atlassian is a diginomica premier partner and funded travel to the event.

What this means for MSP operators

Platforms are starting to manage agents like staff: identity, scoped authority, reviewable results, and a path to revoke. Until every tool a client uses does that, the MSP still keeps the list. Your product is not an Atlassian brochure. It is a per-client agent inventory, then an access review: discover, name an owner, match scope to the job, require approval on send, post, spend, or delete, and drill the revoke.

Channel Insider (Victoria Durgin, Oct 6, 2026) cites IANS Research and Artico Search (113 CISOs, April-May): 66% have a dedicated AI policy, 74% of enterprise AI environments pull data from external sources, and only 29% have adversarially tested their AI systems. The same piece cites Deloitte Q2 2026 CFO Signals: 93% of surveyed North American CFOs use AI across multiple key functions, and 59% say balancing rapid deployment with risk is their biggest AI governance challenge. Corsica Technologies' Peter Rodenhauser told Channel Insider: "We really did try and lead with governance and policy as a conversation." That is the MSP lane: inventory first, then the access review as a recurring line item.

Figures: Consultiply diagram of four modalities plus identity table and CISO stats, and an agent inventory pipeline with cadence and illustrative inventory (not client data). Sources: Atlassian blog; diginomica; Channel Insider.
01

Start with a per-client agent inventory.

List every agent, bot, MCP connection, and service account that can read, write, send, post, or spend. Name an owner for each. If the client cannot list them, that is the first deliverable.

02

Match scope to the job, then gate outbound actions.

Read only, then drafts only, then acts with approval, then narrow acts-alone. Require a human yes on send, post, spend, or delete until the monthly review is clean.

03

Treat coming-soon vendor controls as roadmap, not coverage.

Atlassian's NHI inventory and revoke switch are coming soon. Until they ship and every other client tool has the same, your inventory and kill path are the control.

04

Drill revoke and put time-to-stop in the QBR.

Page path, token revoke, log review, resume one step lower. Time it. Record the result. Policy without a tested stop is a slide, not a service.

05

Label every claim as theirs.

AMP wording, MCP MAU, collaboration volume, and diginomica's diplomatic-protocol framing are Atlassian's or the outlet's. Channel Insider's 29% adversarial-test figure is IANS/Artico via Channel Insider. Your product is the inventory and the monthly access review.

Offer moves this month

01

Deliver a one-page agent inventory per client.

Name, owner, tools touched, credentials in reach, and whether send/post/spend/delete need approval. Ticket gaps in the PSA.

02

Run the access review after the inventory.

Diff permissions, revoke unused tokens, require approval on high-risk actions, and confirm a stop owner exists.

03

Book the monthly AgentOps governance line item.

Week 1 inventory refresh, week 2 permission diff, week 3 revoke drill, week 4 scorecard. Sell it as recurring governance, not a one-time setup.

04

Watch Atlassian NHI inventory and revoke when they ship.

When live, fold them into the same review. Do not wait on one vendor control to start listing agents everywhere else.

Numbers table

Figures below appear in Atlassian's materials, diginomica, or Channel Insider. None were measured by Consultiply.

ClaimWhat was measuredStatus
AMP announce date (Business Wire)Oct 7, 2026 (Team '26 Europe, Amsterdam).Vendor claim
AMP definition (Atlassian blog)Identity, scoped authority, shared context and tasks, and results you can review.Vendor claim
NHI inventory and revoke switch (Atlassian blog)Coming soon. Not described as live.Vendor claim
Human-agent collaboration (Atlassian)Over ten million times a month on the platform.Vendor claim
Atlassian MCP MAU / daily calls (Business Wire)Nearly 2 million monthly active users; over 15 million daily tool calls; 15x in six months.Vendor claim
Token efficiency (Atlassian internal benchmark via BW)Up to 25% fewer tokens for the same Jira/Confluence work on Claude models.Vendor claim
Dedicated AI policy (IANS/Artico via Channel Insider)66% of 113 CISOs (April-May).Vendor claim
Adversarial testing (IANS/Artico via Channel Insider)Only 29% have adversarially tested their AI systems.Vendor claim
External data in AI environments (IANS/Artico via CI)74% pull data from external sources.Vendor claim
CFO AI use / governance tension (Deloitte via CI)93% of surveyed North American CFOs use AI across multiple key functions; 59% cite balancing rapid deployment with risk as biggest AI governance challenge.Vendor claim
Callout

This guide is Consultiply's independent analysis of Atlassian's Team '26 Europe AMP announcement and related coverage, translated for MSP operators. It is not affiliated with or endorsed by Atlassian. The non-human identity inventory and revoke switch are described by Atlassian as coming soon.

Sources

The Atlassian platform is for multiplayer human-agent work (Atlassian blog, Jamil Valliani)

Team '26 Europe - Atlassian's Agentic Multiplayer Protocol sets the scene as agents join the team (diginomica, Phil Wainewright; disclosure: Atlassian premier partner / funded travel)

AI Security Concerns Create a Bigger Opening for MSPs (Channel Insider, Victoria Durgin, Oct 6, 2026)

Where this came from

Reviewed, not run by us.

This guide is Consultiply's independent analysis of Atlassian's Team '26 Europe AMP announcement and related coverage, translated for MSP operators. It is not affiliated with or endorsed by Atlassian. Atlassian blog (Jamil Valliani) and Business Wire press release (Oct 7, 2026) →

Want this as a PDF you can share?

Free to download, no form to fill in.

Get the PDF (~553 KB)