The Problem
Once an organization rolls out an AI coding assistant to hundreds of engineers through a central LLM gateway, a new operational load appears: onboarding requests, login issues, expired invitation links, requests to join internal teams. Each one traditionally needs a manual permission check, user creation in the system, and follow-up on internal approval workflows.
The Solution
The team built an access-management agent that handles the full lifecycle of employee access to enterprise AI tools directly from Jira tickets:
- Creates new users in the gateway, assigns them to the correct team, and sends onboarding emails.
- Reissues invitation links for users whose links have expired.
- Adds existing users to additional teams when they need more access.
- Responds to known bugs with an immediate workaround, so the user doesn’t wait for a support engineer.
Built-in Safety Layer: Human Approval in the Right Place
Not every action should be autonomous. Creating a new user requires explicit approval from the relevant organizational authority. If the requester isn’t authorized, the agent tags the right approver on the ticket and proceeds only after confirmation. Lower-risk actions, like reissuing an invitation link, are handled immediately.
How It Works in Practice
- A ticket arrives in the AI-tool access category.
- The agent identifies the request type: new user, re-invitation, team join, or known bug.
- If approval is required, it tags the approver and waits for a reply on the ticket.
- After approval – or immediately if none is needed – it retrieves the required credentials from the organization’s secrets store, provisions the user, sends notifications, and updates the ticket.
The Results
Onboarding that previously needed a support engineer at every step is now a near-autonomous end-to-end flow, with human-in-the-loop governance exactly where judgment is needed. Wait times for new employees seeking AI tools dropped from days to hours.
Why It’s Worth Talking About
Modern AI can execute complete authorization workflows, not just answer questions – knowing when to stop for a human and when to act alone, while managing access to the AI tools themselves.
