Release assistant
Registered- Responsible
- Platform team
- Purpose
- Release readiness context
Keep authority clear across your AI stack.
Bring the agents you run into view. Give them the right organizational context, define their boundaries, and keep the decisions on record.
Know who runs each agent and who is responsible.
Bring registered agents and discovered activity into one review. See which agents have an owner, which need a decision, and where their activity was observed.
Claim discovered automation deliberately, with an owner and rules. Connected evidence defines what is visible; it does not establish a complete inventory of your business.
Explore agent oversightAgent reviewOwnership and status
Support workflow
Discovered- Responsible
- Unassigned
- Source
- Imported activity
- Next step
- Review ownership and rules
Knowledge assistant
Registered- Responsible
- IT team
- Purpose
- Internal knowledge retrieval
GitHub
Jira
Slack
DriveAgent briefing · Permitted sources
Enough context to ask the right question.
- Commitment
- Release scheduled for September 18.
- Open work
- Change review still needs a decision.
- Evidence
- Release plan and review record, with source references.
- Next question
- Can the review finish before the release?
Give agents context within their access boundaries.
Connect the systems where work happens. Let agents request relevant commitments, risks, and status through permission-filtered memory and briefings.
Choose the source scope at connection, then respect each agent’s access. Secrets and private member data stay outside shared context. Reading the record does not grant authority to change a system.
Explore organizational memoryUnderstand a rule’s effect as you roll it out.
Where staged rollout is enabled, observe a rule’s behavior, review its recommendations, and choose when to enforce it. Each stage change stays on record.
Release assistant rule
Production configuration changes require approval.
Observing
See what the rule would do.
Record the rule’s would-be outcome as configuration requests arrive. Review its behavior before this rule starts holding changes.
Change requestedPayment service rate limit
Would require approval
This rule records its would-be hold. It does not enforce it at this stage.
- Review
- Would-be outcome recorded
- Effect
- Not enforced by this rule
Other permissions, policies, and operating limits apply at every stage.
Know where you have visibility and control.
Keep the connection path attached to the evidence. An imported event and a decision checked before execution tell you different things.
A check before the action proceeds.
Route supported requests through the governed path. Permissions, policies, evidence, and operating limits determine whether the action can proceed.
RequestGovernance checksRecorded verdict
A record of activity that already happened.
Import external agent activity, with Azure AI Foundry as the verified source. The record preserves its source and observed status; it cannot retroactively block the action.
External activityImported logObserved evidence
Pause with a record.
Hold governed actions for an agent or the workspace while you investigate. Keep the pause, resume, and attempted actions available for review.
Carry the history forward.
Trace decisions to their actors, rules, and receipts. Prepare scoped evidence with coverage gaps visible, so reviews start from what the record supports.
Have questions?
Do we need to replace our agents or AI tools?
No. Register and govern agents you already run in your own tools and environments. Connect supported sources for organizational context and route supported actions through the governed path. A source connection or imported log does not, by itself, place an external agent under enforcement.
Can we discover agents that have not been registered?
Where discovery and supported activity imports are enabled, evidence can surface unregistered automation. Discovered agents remain separate from governed agents until an authorized person claims them, assigns ownership, and establishes rules. Coverage depends on the sources connected; discovery is not a complete scan of every agent or device in the business.
How much organizational context can an agent access?
Memory retrieval and briefings are filtered by the agent’s permissions. Briefings contain a bounded selection of relevant commitments, risks, status, and facts, with sources and timing attached. Secrets and private member-vault data are excluded. Access to context does not grant authority to perform an action.
Can we review a rule before it starts holding actions?
Where staged rule rollout is enabled, an agent rule can observe activity, surface its would-be decisions for review, and then move into enforcement. An authorized person changes the stage, and that change leaves a receipt. Observing and recommending do not apply that rule’s hold; other policy checks, permissions, and operating limits remain in force.
What is the difference between an imported log and an enforced action?
An imported log records activity that has already happened. It provides observed evidence and cannot prove that execution was blocked or approved beforehand. An action routed through the enforcement path is evaluated before it proceeds. The recorded evidence distinguishes those paths so a review can state the actual coverage.
Can we pause or retire an agent?
Authorized operators can pause a registered agent or the workspace’s governed actions. Attempts through the governed path are held while the pause applies, and the pause and resume remain on record. Where decommissioning is enabled, retirement revokes governed authority while preserving history. External runtimes and upstream credentials still need their own offboarding steps.
Does connecting a business system grant write access?
No. Source connections ingest evidence within the scope you authorize. They do not automatically authorize changes in the connected system. Supported actions require their own permissions, policy checks, and any required confirmation or approval.
Define how your AI stack should operate.
Connect the context, set the boundaries,
and make the next decision traceable.