Salesforce Will Soon Charge When Your AI Agents Call It

 Salesforce has confirmed that it will start charging for the calls AI agents make into its platform. Its own help documentation says every successful call a registered agent makes to Salesforce, over the Model Context Protocol or a direct API, will be metered as a Headless Platform Interaction and paid for in Flex Credits. The charge is not live yet, and Salesforce has not published what each interaction will cost.

A confirmed meter with an unpublished rate is the combination buyers find hardest to plan around. This article sets out what is known, what is not, and what is worth doing before the meter starts.

What Salesforce has announced

The change sits under the platform layer that Salesforce’s documentation calls AIforce. It was first announced at TDX 2026 as Headless 360, and on 9 October Marc Benioff said it is now officially called SIforce, its third name in about six months. The layer lets agents work with Salesforce without a browser.

Alongside it comes a requirement for agents to be registered. Under the new Agentic Identity framework, each agent gets its own credentials and permissions instead of acting under the login of the person it assists. Usage is then tracked in Digital Wallet under a new usage type on the Flex Credits rate card.

Two protections are built in. Salesforce says metering applies only in active production orgs, so sandboxes, scratch orgs and Developer Edition orgs do not consume credits. Development and testing should therefore not draw down your balance. Salesforce has also promised 30 days’ notice before metering starts, once the multiplier that sets how many Flex Credits each interaction uses has been added to the rate card.

Salesforce has not given a start date. Because of the notice commitment, billing cannot begin until the multiplier is published and a month has passed.

What is not known

The multiplier is listed as to be announced, so nobody can yet price an agent session. For reference, Flex Credits are sold at 500 dollars per 100,000 credits, and a standard Agentforce action uses 20 credits, which works out at about ten cents. If an interaction carried a similar weight the arithmetic would be easy, but Salesforce has not said that it will.

The number of interactions also depends on how an agent is built. A Salesforce spokesperson told Salesforce Ben that the cost per interaction does not vary, while the number of interactions charged depends on the agent’s behaviour and implementation. Two agents doing the same job can therefore generate very different bills, depending on how many calls each one makes to get the work done.

Who is affected

The model is not limited to Salesforce’s own Agentforce. It also covers outside tools that call Salesforce, such as Claude or ChatGPT, and it applies to agentic API traffic as well as to MCP. Organisations that bring these tools in will already be paying the AI provider, and Flex Credit consumption adds a second cost on the Salesforce side. The entry point is easy to reach. Salesforce’s developer documentation describes a Headless 360 MCP server, in beta since July 2026, that gives an MCP-aware agent access across Salesforce through a single connection.

Traditional integrations are meant to stay as they are. Salesforce says existing API-based integrations continue under their current pricing and security arrangements, and that the new model is tied to registered agentic workflows. The boundary between the two is the part to clarify. If an existing integration starts using an AI model to decide what to call, is it still a traditional integration? Salesforce says it identifies and isolates third-party agent traffic once an agent is registered, but customers will want to know exactly how that classification works for their own traffic.

Where the credits come from

These interactions will draw on the same Flex Credits pool as other agent usage. Salesforce’s new editions include annual allocations of 500,000 credits for Core, one million for Advanced and 2.75 million for Max. Organisations that sized those allocations around conversations and actions in Agentforce should assume some of that capacity will now go to agent calls into the platform. That could bring forward the point at which they need to buy more credits.

What to do before the meter starts

Start with an inventory. List every AI tool, assistant and automation that connects to Salesforce, and note whether each one uses MCP, the API or a packaged connector. Work out which would count as registered agents and which would remain traditional integrations. Then ask Salesforce to confirm that classification in writing for your estate.

Next, ask for numbers. Request the multiplier as soon as it exists, and ask whether Salesforce will fix it for the length of your contract. Model low, expected and high usage, because the same agent can behave differently as prompts and tools change. Ask whether your existing credits can be used for these calls, and what happens if the pool runs out part way through the term.

Finally, think about design. Because the number of calls drives the cost, an agent that fetches data efficiently will cost less than one that makes many small calls. Salesforce says its tooling includes search designed to reduce token consumption on the model side. Whether that also reduces the number of metered interactions is a separate question, and one worth testing.

A note on names

Because the platform layer has been renamed twice since April, documentation, release notes and sales conversations may use different names for the same thing. Search for all three, Headless 360, AIforce and SIforce, when you look for guidance. In your contract and order documents, ask for the language to describe the capability itself, such as metered calls from registered agents, rather than a brand name that may change again.

Conclusion

Salesforce has confirmed that calls from registered AI agents will consume Flex Credits, with the rate to follow and 30 days’ notice before billing starts. Until the multiplier is published, the cost cannot be calculated, and it will depend as much on how agents are built as on how many you run.

Use the gap to prepare. Know which agents and tools touch Salesforce, get the classification rules confirmed in writing, and try to secure a fixed rate for your contract term before the meter goes live.

 

More on the Blog