
AI is billed by the call
Every request an AI agent sends to a model such as ChatGPT, Claude or Gemini is metered and billed. An agent can send call after call, all day, and the total shows up on the invoice, after the money is spent.
- Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, citing escalating costs, unclear business value or inadequate risk controls.¹
- Gartner also predicts AI inference costs per agentic workflow will rise more than five times through 2028, as advanced reasoning models typically use five to 30 times more tokens than basic chatbots.²
- 98% of State of FinOps 2026 respondents now manage AI spend, up from 31% two years earlier.³
A dashboard tells you what was spent. A budget has to act before the money is gone.
Two ways the bill runs away
- Nobody controls the calls. People and their agents call AI models with their own keys, the spend lands on the company, and no one owns it or limits it.
- An agent gets stuck in a loop. It sends the same paid call again and again: the same tool call with the same result, or the same error retried.
Either way the result is a bill nobody approved. Security cannot say who owned the spend, and finance can freeze the project. See budget and token spend and AI visibility.
The one idea: check the budget before the call is sent
Argent Strategic Intelligence sits between your agents and the AI models they call, so the spend limit is enforced where the call is made, before the money is spent.
- Agents call models through Argent with an Argent key. The company's AI provider key stays in its vault and Argent attaches it to each call on the way out, so by default an agent never holds it. See Keyless Agents.
- Budgets follow the org chart: company, organization, department, person and agent, in dollars, for a day, a week or a month.
- Every refusal, budget change and approved raise is a signed record. See One-Click Audit Evidence.
How it works, step by step
- Fund. An admin sets budgets down the org chart on the Budgets page. You can only hand out what you hold: the budgets under a level never add up to more than it has.
- Turn enforcement on. The budget enforcement switch ships off; until it is on, spend is counted and nothing is stopped. Once it is on, an agent nobody has funded cannot spend.
- Price the call. Each request is priced before it is sent, at the most it could cost: the prompt plus the most output it may ask for, at the model's catalog price. By default, a model with no price cannot spend against a budget.
- Hold it at every level. That amount is held against every funded budget above the agent, from its own budget up to the company, so many calls at once cannot slip past the cap.
- Allow or refuse. If every level can cover it, the call goes through. If any level cannot, it is refused: nothing reaches the model, and the reason names the level that ran dry.
- Settle. When the answer comes back, the hold is trued up to what the call actually cost, and the signed record carries that cost.
What the operator sees
| Where | What it shows |
|---|---|
| Budgets page | Four tabs: Overview (a spreadsheet-style sheet to plan and allocate), Spend, Approvals and Caps. A banner reads ENFORCED: "Requests past their budget are denied." |
| A refusal | A plain reason, such as "the department budget is exhausted for this period", on a signed record. |
| Owner portal | A person sees their funding chain, splits their budget across their own agents, and asks for more. |
| Approvals | Each request with who asked, the ceiling now, the ceiling if approved, when the money runs out and when it takes effect. |
| Caps | Emergency caps, active and history, stamped TRIPPED when a cap is hit. |
| Spend | Spend by company, organization, department, owner, agent, model and provider, with top spenders. |
| Alerts | Notices before the money runs out, sent to the notification channels you choose. |
Every raise is a person's decision
An agent never raises its own limit. A person asks, and someone who holds the money decides.
- Ask. The owner chooses Request more in Argent and enters how much more and why. The request funds nothing by itself.
- Decide. An approver who holds the money, such as the department head, sees what approval does to the ceiling and when the money would run out, then approves for now or for next period. To deny, they write a reason, and no money moves. Nobody approves their own request.
- Ticket it. Connect ServiceNow or Jira Service Management and Argent can open a ticket for every new request and push the decision onto it. An approval made on the Jira Service Management ticket flows back into Argent and is applied with the console's own checks.
More controls on the same budgets:
- Shared pool. An owner or an admin can let an agent borrow from the money its owner has not handed out.
- Emergency cap. An operator brakes spend with a dollar limit over a rolling hour, day or week on any agent, person, department, organization or the company.
- Hard stop. Nothing under a budget can borrow from money above it.
- Carry-over. Unused money can be added to the next period, up to one period's allotment.
- Cheaper model near the limit. A budget can switch an agent to a cheaper approved model as it nears its limit, so the work keeps going for less. See Model Control.
Loops: stop the waste before the cap has to
Every repeat from a stuck agent is a paid call.
- Argent's loop detector spots an agent repeating the same step: the same tool call with the same result, two steps back and forth, or the same error again and again. By default it records each finding.
- When you set loop detection to enforce, Argent refuses the repeating request while the agent's budget still has money. The request never reaches the model, so the spend that would have followed stops.
- The Loops page lists the finding. An admin opens the agent's page and quarantines it: every request from it is refused until it is released, and nothing is deleted. See Agent Behavior Watch.
Where it fits and how to start
| For | Why spend caps matter |
|---|---|
| Financial services and insurance | When fraud, claims and service teams run agents, each department gets its own ceiling and an owner who answers for it. |
| Retail and healthcare | Customer-facing and back-office assistants share one budget tree, so one busy agent cannot spend another team's money. |
| Public sector | Spending follows approved allotments, and every raise is a named person's decision on the record. |
| Your goals | Budget and token spend, audit and accountability, agent identity and ownership, compliance evidence. |
| Use cases | Securing AI coding agents, rolling out copilots and chat, customer-facing AI assistants. |
- Fund first. Set budgets while enforcement is off: spend is counted and nothing stops.
- Watch, then enforce. Turn budget enforcement on with the cost area in watch mode, and Argent records each call a budget would have refused while the call goes through; emergency caps still apply. Switch to enforce and every call must fit every budget above its agent. See Watch Before You Enforce.
- Run it where you need it. Air-gapped and on-premises, private cloud or your public cloud VPC. See all industries and all capabilities.