The first week of October gave project managers something more interesting than another AI assistant promising to tidy up their to-do lists. AI agents are getting dangerously close to running the work themselves. Now everyone wants to own the map they follow.
At Team ’26 Europe on October 6, Atlassian introduced its Agentic Multiplayer Protocol (AMP), designed to give agents identities, clear permissions, and reviewable actions. A day later, Wrike unveiled plans for autonomous project work backed by its Work Intelligence Graph. Chief Product Officer Alexey Korotich explained the problem:
“Most enterprise AI fails because it operates in a vacuum, generating content without understanding how work actually gets strategized, planned, and delivered.”
Google piled in on October 8 with a Gemini agent designed to coordinate work across business systems. It's still in private preview, so nobody should hand it the project plan and head home just yet. But with OpenAI's Dots and Space pursuing the same opportunity, things are definitely heating up.
Here's the question. If agents can work from ChatGPT, Slack, or Teams, who controls the project record underneath? Who owns the overdue task, which deadlines depend on it, and who approved the agent's changes?
SAP seems interested in that answer too. On October 6, it agreed to acquire TechWolf and its "context graph for work."
Meanwhile, Gartner's October 5 Collaborative Work Management research now evaluates agentic AI alongside planning and governance. Seems like the industry's getting serious about AI managing actual projects.
Suddenly, those boring old task boards look rather valuable.
Why Are Work Context Graphs Suddenly So Important for AI Agents?
You can stare at a Jira ticket for ten minutes and still have no idea why its deadline changed. Someone agreed to move a launch in Tuesday's meeting, the decision never made it into the ticket, and now three teams are working to different dates. Ask an AI agent to sort out the plan, and you might just get the confusion reorganized for you.
Hence the growing value of work context graphs. They connect tasks to their owners, dependencies, and decision history. Gartner's September 15 research suggests that context could help agents make fewer mistakes.
At Team ’26 Europe, Atlassian CEO Mike Cannon-Brookes put it this way:
“What won’t change are the fundamentals of great teamwork. Shared context, clear ownership, and end-to-end visibility.”
Atlassian says its Teamwork Graph contains over 250 billion connections. In internal tests, agents answered questions up to 44% more accurately while using 48% fewer tokens. I'd rather see what happens when an agent discovers that moving a launch leaves the testing team with two days' work and one day to do it. There’s also the Atlassian expanded MCP offering that includes over 220 tools and reportedly handles 15 million calls daily. That's considerable access to live projects.
Right now, the Tempo 2026 survey is saying that 91% of 300 senior project and portfolio leaders were using or piloting AI, yet just 33% were delegating delivery work. Attribution problems affected 55% of teams without production agents, compared with 16% using them.
Glean's 2026 Work AI Index adds another frustration: workers reported spending 6.4 hours weekly supplying missing context and fixing AI mistakes. That's nearly a working day spent explaining things to your new assistant. No wonder context graphs are gaining attention.
How Do Asana’s Work Graph and Atlassian’s Teamwork Graph Compare?
Asana's Work Graph gives employees and AI Teammates a shared understanding of projects. Atlassian is making Teamwork Graph available as a context source for Rovo and external agents to query. The difference comes down to how each company expects agents to use its platform, including whether those agents operate inside the product or draw on its information from elsewhere.
A small October 8 update brought more impact to the context story. Asana can now roll subtask dates up into a parent task, but the action is manual and can wipe out planned buffer time. Anyone who's watched a carefully arranged project schedule unravel over one innocent date change will appreciate the problem.
Atlassian's approach is a little different. Its expanded MCP server and CLI let outside assistants work across Jira and Confluence, using Teamwork Graph for context. AMP introduces clearer attribution and scoped permissions, although dedicated agent accounts and a complete identity inventory are still coming. Both vendors support external AI connections, but Atlassian is pushing harder to make its graph useful wherever agents operate.
Elsewhere, monday.com has its boards and cross-board relationships, Linear Loops responds to project changes, and Notion and Adobe Workfront are opening live work to outside agents.
Atlassian's October updates also bring Jira capacity and budget signals, portfolio staffing scenarios, and agent-drafted progress reports requiring human approval.
Personally, I'd be more interested in whether an agent can tell me that moving Friday's deadline will wreck next month's launch. Knowing everything about a task is marvelous; it’s better to know what happens when someone changes it.
Can ChatGPT Space Replace Asana, Jira or monday.com?
There's something very appealing about telling an AI agent to sort out a project and having it actually get on with it, especially when the alternative is another meeting about why nobody's updated Jira.
OpenAI has spent September giving agents more responsibility. Team Tasks can handle recurring project updates, tracking progress, blockers, and next steps as part of the workspace-agent approach we've been following. There's a practical complication, though. Retiring the model behind a workflow won't make Monday's reporting deadline disappear.
Space has shared Pages, configurable permissions, Activity View, and approval rules. Unfortunately, automatic Page updates through “Keep updated” aren't available at launch. That's a bit of a tricky limitation for something promising to keep up with moving projects.
I wouldn't cancel the Jira licenses just yet.
Google's October 8 Gemini agent promises persistent context and dedicated identities across applications, although it's still in private preview. Meta's new Muse integrations also bring Asana and Notion into another assistant's reach. Meanwhile, Atlassian's OpenAI partnership shows the supposed rivals are happy to work together.
Space is available on Pro, Business and Enterprise, with Dots more restricted, including an Enterprise beta.
My suspicion? We'll increasingly ask agents to manage work from whichever app we're already using. But when a launch goes wrong, someone will still need to explain the missed dependencies and questionable decisions.
And I suspect that someone will be opening Jira or Asana. It's the Slack and Salesforce situation all over again.
What Should Buyers Ask Before Trusting AI Agent Context?
As AI governance concerns continue to fester, there’s a lot to consider here. I'd like to see an agent tackle a project that's already going wrong, since, let’s be honest, that happens pretty regularly. Move a launch forward two weeks, throw in an unavailable developer, and see whether it notices the trouble you've just caused.
Here are five questions I'd ask before giving it access to anything important.
1. What does the agent actually know?
Your graph might contain thousands of connections and still miss Friday's conversation with the sponsor. PMI made that point: project managers often sense changing priorities before they're documented. Ask how quickly the graph catches up, and what happens when records disagree.
2. What's a useful result going to cost?
