Salesforce Is Changing the Rules for AI Agents [And Many People Haven't Realized It Yet]

Salesforce is preparing an important change in how artificial intelligence agents interact with its platform. And I believe everyone working in the Salesforce ecosystem should be paying attention.

Not just those using Agentforce. I'm also talking about developers, architects, ISVs, consulting partners, and companies building AI agents using external technologies.

Because we're entering a new era where APIs will no longer be used exclusively by traditional applications or system integrations. Increasingly, they'll be used by autonomous agents capable of reasoning, making decisions, and executing tasks. And Salesforce is preparing its platform for that reality.

What Salesforce Has Announced

So what exactly has Salesforce announced? In an official Salesforce Help article titled "Understand AIforce Impact," the company outlines upcoming changes involving AIforce and the Headless Toolkit. These changes include new security capabilities, agent registration, and a consumption-based billing model.

Agentic Identity

One of the key concepts is Agentic Identity. The idea is to give AI agents their own identity within Salesforce, rather than having them operate under the identity of the human they're assisting. This makes it possible to manage credentials, define more granular permissions, and improve oversight of agent activities.

From an architecture and security perspective, I see this as an important step forward. As agents become more autonomous, we need better ways to govern what they can and cannot do.

Headless Platform Interactions

But there's another change worth paying close attention to: Headless Platform Interactions. Salesforce has announced that successful calls made by registered agents, whether through MCP or direct APIs, will be counted as Headless Platform Interactions (HPIs). These interactions will be associated with Flex Credits consumption and can be monitored through Digital Wallet.

There's an important distinction here. Salesforce has not yet announced the final Flex Credits multiplier for these interactions. According to its official documentation, agentic consumption under this new model is not yet being metered for billing. Salesforce also states that it will provide 30 days' notice before metering begins.

That means we still can't determine the exact cost of an individual interaction. And it would be incorrect to claim that all Salesforce API calls will suddenly incur additional charges. Salesforce explicitly states that traditional integrations retain their existing pricing and security arrangements. The new model specifically targets agentic interactions covered by AIforce and the Headless Toolkit.

Why This Matters

Why does this matter so much? Because it introduces a variable that many AI agent developers haven't previously needed to consider in this way.

Imagine a sales agent interacting with customers through WhatsApp. The agent might query Salesforce to identify a customer, retrieve products, check commercial information, update records, or create orders. Completing a single task could require multiple calls to the platform. If those operations are performed by a registered agent and fall within the new HPI model, the number of interactions becomes a relevant factor when estimating consumption.

This means that, in addition to accounting for language model costs, infrastructure, and automation, we'll also need to evaluate interactions with the underlying enterprise platform. That's not necessarily a bad thing. In fact, it could encourage better architectural practices, observability, and optimization. But it does introduce another consideration when designing enterprise solutions powered by autonomous agents.

There's also an interesting opportunity for Salesforce architects. For years, a significant part of our work has involved designing efficient, secure, and scalable integrations. Now we'll need to apply those same principles to systems where execution decisions may be made by AI agents.

And that raises some interesting questions:

• How many API calls does an agent actually need to complete a task?
• How do we design tools and operations that interact efficiently with Salesforce?
• How do we manage permissions for agents capable of executing multiple business processes?
• How do we monitor their activity?
• How do we forecast operational costs when the number of interactions may vary from one execution to another?

I believe these questions will become increasingly important in conversations among customers, partners, and development teams.

Transition Conditions

Another important detail: Salesforce has outlined different transition conditions. According to the official documentation:

• New customers must register agents using APIs within three months of being notified that Agentic Identity is available.
• For existing customers using agents through APIs, Salesforce establishes migration to the new model at renewal.
• For customers using Salesforce MCP servers, the transition period is three months from the Agentic Identity availability notification, regardless of whether they are new or existing customers.

Salesforce also clarifies that HPI consumption applies only to active production organizations. Sandboxes, scratch orgs, and Developer Edition organizations will not consume Flex Credits for HPIs under the announced model. This is particularly relevant for those of us who develop, test, and distribute software on Salesforce.

What This Means for ISVs and Consulting Partners

What does all this mean for ISVs and consulting partners? I believe we have an opportunity to help customers navigate a transition that goes far beyond adding a chatbot or connecting a language model. We're moving from integrating applications to integrating autonomous agents. And that requires us to rethink identities, permissions, security, architecture, monitoring, and consumption models.

It will also be important to understand how these capabilities fit into Salesforce's different licensing and commercial arrangements. For companies building products on the platform, these details will be essential before committing to new architectures or pricing models.

There's something about this evolution that I find particularly encouraging. Salesforce is recognizing that AI agents need a different way to interact with enterprise platforms. Simply exposing an API is no longer enough. When we're talking about autonomous agents, we need to know which agent is executing an operation, what permissions it has, and how its activities can be monitored and governed.

The move toward dedicated agent identities is especially interesting. And the consumption model opens an equally important conversation about balancing innovation, cost predictability, and platform sustainability.

One of Salesforce's greatest opportunities is to become a central enterprise platform where AI agents built with different technologies and from different providers can operate securely. After more than 30 years building enterprise software, there's one thing I'm convinced of: The value of a platform isn't determined solely by the features it offers. It's also determined by everything its ecosystem can build on top of it. And Salesforce has an extraordinary ecosystem of developers, architects, partners, and companies that have spent years extending its capabilities.

That's why I believe this evolution deserves an open, technical, and constructive conversation. There are still commercial details to be clarified, particularly the final Flex Credits multiplier for HPIs. In the meantime, we have an opportunity to prepare, revisit our architectures, and understand what this next chapter will look like.

I'd love to hear from others in the Salesforce ecosystem. How are you preparing your architectures to integrate AI agents? Do you think Agentic Identity and the new consumption model will change how we design integrations? And what additional information would help you accurately forecast the costs of your solutions?

Looking forward to hearing your thoughts.

Official source: Salesforce Help - Understand AIforce Impact (September 17, 2026): https://help.salesforce.com/s/articleView?id=005360285&type=1