# Agencies Reevaluate AI Agent Costs

The signal warns that government AI agents running continuously may not fit budgets structured around user seats and software licenses, creating affordability and funding-control risks as usage grows. It also flags August ADA ceilings as a potential constraint on successful programs, though it does not identify an agency, procurement, or specific ceiling. For procurement teams, the central recommendation is to assess AI on cost per unit of work—not token consumption alone—and account for ongoing agent usage when planning budgets and contract pricing.

- Agencies and contractors should develop cost-per-task or cost-per-output measures that reflect continuous agent use, rather than relying only on seat or license counts.
- When structuring AI procurements, align pricing and budget controls with expected workload so increased usage does not create unplanned costs or funding disputes.

**Jurisdictions:** federal
**Industries:** Information Technology
**Topics:** Artificial Intelligence
**Published:** November 11, 2026

### Sources
- [The more govt AI works, the more likely the govt turns it off.

Budgets built for seats & licenses break with 24/7 agents. ADA ceilings cap the winners in August. Tokens aren’t the cost; measure cost per unit of work this year or the fights get worse. https://t.co/sXRVL9S8wi](https://x.com/alexcohenof1/status/2120626996862456283) - twitter-fed-procurement