What the assistants can see, and why that is a permission question.
The Teamwork Graph is the connected representation of work, people, and content across the Atlassian estate, and it is what assistive and agent capability reads from. Our work is understanding what assistive capability can reach, and governing it at the identity layer.
The Teamwork Graph is the connected model of work, people, and content that everything assistive reads from. Its reach is decided by the permission model underneath it.
What it is
The Teamwork Graph is the connected representation of work, people, and content across the Atlassian estate, and it is what assistive and agent capability reads from.
It matters to anyone being asked what their assistants can actually see, which is increasingly a question from risk rather than from engineering.
What it does
It connects work across products
Which is what makes assistive answers useful, and also what makes their reach worth understanding.
It reflects existing permissions
It does not grant access. It surfaces what was already granted, including what was granted by accident.
Where it fits
L4, agents and automation, governed entirely from L1. The graph does not decide access; it reflects it. Everything reading from it inherits your permission model, which makes that model the only real control.
What we do with it
Permission review at the source
What the graph can reach is what your groups permit, so the audit is an identity audit.
Policy for what reads from it
Which assistants and agents may connect, and what evidence is retained.
Reviewing it as an operation
Access changes continuously, so what the graph reflects changes with it. That is a recurring review rather than a one-off audit.
Where it earns its place
An organization answering a risk questionnaire
The question is what assistive capability can see, and the answer is a permission model rather than a product description.
An estate with a decade of accumulated access
The graph makes the accumulation legible, which is uncomfortable and useful.
Proof
Of security leaders say they lack full visibility into the machine identities already running in their environment.
2026 CISO AI Risk Report, 235 large-enterprise leaders.
Questions we get
Can we limit what it indexes?
Licensing
The graph is part of the platform rather than a separate purchase, though edition affects what reads from it. That reconciliation sits with the rest of the estate in licensing and renewals.