Enterprise business application vendors are setting the terms for enterprise AI agents. They determine what you pay, what agents can access, and how you monitor and control them. What may be affordable in a pilot can become very expensive as charges accumulate quickly across the end-to-end business process. By the time you realize what has happened, the agents are embedded in your business. CIOs and heads of enterprise business applications need to negotiate the terms before adoption turns into dependency.

The Meter, The Gate, And The Referee Are Converging

New pricing models are emerging, moving away from per-license pricing to a meter on agent activities. Salesforce Agentforce pricing lists some actions at $0.10, while SAP’s AI commercial model assigns 0.02 AI Units to certain agent actions. Workday AI Flex Credits price completed tasks at five times the rate of information retrieval. The models differ, but the commercial direction is consistent: As agent activity grows, vendor consumption revenue grows with it. A low per-action price can quickly escalate costs across your business processes.

Accessing your enterprise business data just got a lot more difficult. SAP’s April 2026 API policy analysis describes restrictions on third-party agents planning or executing sequences of SAP API calls outside approved paths. In practice, an external finance agent may need to reach an SAP transaction through a Joule-mediated route. SAP has a legitimate interest in protecting application integrity. But when the required route is proprietary and billable, an API policy becomes an architecture and negotiating decision.

As your organization merges multiple products across your technology stack, who controls how the AI agents interact with other systems is important. Application and platform vendors increasingly offer to discover agents, enforce policies, monitor activity, and govern interactions across vendor boundaries. Enterprises will need many of these capabilities. But introductory pricing is not a control strategy.

Together, these moves put the meter, the gate, and the referee in one vendor’s hands. Concentration may improve control, but it also allows the vendor to price the work, govern its execution, and provide the evidence of its value. Once hundreds of agents depend on that platform, the enterprise may lose the ability to challenge the bill or change the terms without disrupting operations.

Know Your Rights Before Agents Execute

Before CIOs and heads of enterprise business applications turn over all control to a single provider, they should understand the agent cost impact end-to-end for critical business processes. For example, consider an invoice agent. It starts by summarizing a discrepancy. Soon, it will match the invoice to a purchase order, resolve exceptions, update the ERP system, and prepare payment. The agent does not suddenly become more intelligent. It becomes more connected and more authorized. That expanded capability comes at a price.

The enterprise’s control should rise with its authority. Analyze your core business processes with these six key lenses.

  1. Know who acted. “The AI did it” is not an acceptable answer. Every production agent needs its own identity, owner, and bounded authority. If you cannot connect an action to the agent, initiator, source, and policy, the agent is not ready to execute it.
  2. Keep the evidence. A transaction log will not tell the whole story. Preserve what the agent saw, which rules it applied, which tools it called, and what it changed. If the complete record lives only in the vendor’s console, you have visibility on the vendor’s terms.
  3. Verify the value. A completed action is not a completed business outcome. Compare agent consumption with completed transactions, exceptions, human intervention, corrections, reversals, cycle time, and financial impact. Retain the telemetry needed to reconcile charges and dispute consumption. If you cannot verify the outcome independently, you are measuring activity, not value.
  4. Know the real price. Eventually, the business will stop buying agents by the task and start buying them by the outcome. Include retrieval, API calls, handoffs, retries, approvals, exceptions, and reversals. If the business case works only at pilot volume or promotional rates, it is not a production business case.
  5. Choose the route. Vendors should protect their applications, but they should not quietly dictate every intermediary in the process. Define supported routes, mandatory gateways, charges, and telemetry rights in the contract. What looks like an API condition today can become renewal leverage tomorrow.
  6. Keep the off switch. Disabling the interface may not stop credentials, schedules, connectors, or downstream agents. Test whether you can revoke authority, halt execution, preserve evidence, and find unfinished work. If you cannot stop the agent independently, you do not fully control it.

These rights are easiest to establish while the agent remains optional. Once people and processes rely on it, changing the terms becomes much harder.

Pay Attention To What Changes

Once an agent moves into production, you must continue to pay attention to how it works, what work it does, and how well it is governed.

Watch the price of execution. Retrieving information and completing work are not the same thing, and vendors may increasingly price them differently. Before scaling an agent, calculate the cost of the complete process, including retries, approvals, exceptions, and reversals.

Watch the route. A vendor may require outside agents to use its gateway, agent, or control plane. Some restrictions are necessary to protect the application. Others can create another charge and another dependency. Ask what remains supported if you choose a different agent or orchestration platform.

Watch the governance package. The agent inventory may be included while telemetry exports, evidence retention, policy enforcement, or shutdown controls require another license. Establish what the production environment will cost, not just what the pilot includes.

Set The Terms Before Agent Dependency

The risk is not simply that AI agents become expensive. It is that costs and dependencies grow together, leaving the enterprise with less control over both. As agents move from helping employees to executing business processes, the terms governing pricing, access, and oversight become harder to change.

Before scaling, CIOs and application leaders need to establish the full cost of a completed process, define what agents can do without approval, and retain the evidence and authority to stop them. Negotiate those rights while alternatives are still practical, not after everyday operations depend on the vendor’s platform.

Adopt agents for the value they deliver, but do not let adoption decide the terms for you.

Let’s Connect

Forrester clients can schedule an inquiry or guidance session to work through a specific agent use case and the decisions that should be settled before production.

Share