Built with Mobileye’s cloud support team
The Problem
The organization’s central internal AI platform runs on monthly token quotas assigned to budget groups. When a group exhausts its allocation, its users are blocked. Quota-increase requests look simple but involve layered policy: distinguishing groups that are truly blocked from those with remaining balance, different caps by quota tier, and emergency rules such as limits on increases per month.
Calculating this by hand for every request is slow and error-prone.
The Solution
The quota agent automates the entire decision tree:
- Identifies the relevant budget group from the ticket or by querying the user directory.
- Retrieves real-time quota data from the platform’s API: current usage, available balance, extra quota already granted this month, block status.
- Reconstructs the group’s original monthly quota, avoiding the classic mistake of stacking increases on top of already-elevated limits.
- Applies scenario-based policy:
- Not actually blocked? The agent doesn’t increase anything – it asks the user why they need more, since they may not know they still have balance.
- Blocked, small group? Raised directly up to the tier ceiling.
- Blocked, large group, no increase yet this month? A percentage-based increase.
- Blocked after already receiving an increase this month? An immediate emergency partial release to unblock the user now, while the budget owner is tagged on the ticket for approval of anything further.
- Executes the change via the quota API and updates the ticket.
The Results
Users blocked from the organization’s central AI tool get immediate relief, even in emergency scenarios. Budget policy is applied with 100% consistency and full governance documentation, and management is involved exactly when needed.
Why It’s Worth Talking About
Precise real-time business accounting by an AI agent: external API calls, arithmetic, multi-layered policy, and intelligent escalation checkpoints for human decision-makers.
