Marc Benioff posted a short message on X: AIForce was officially being renamed SIForce. The whole ecosystem asked the same question: new platform or new name.
The short answer is that, so far, SIForce is AIforce under a different name. There is no new architecture, no new API and no newly published price that says otherwise.
But the rebrand hides a much more interesting story. And that story does affect you, especially if your company already connects AI agents to the market-leading CRM platform.
The third name in six months
This same initiative has had three names in a little over half a year.
First it was Headless 360: the idea of exposing all of the platform's capabilities through APIs, MCP, CLI and skills. Then the Release Notes recorded the change to AIforce, which was formally introduced at Dreamforce.
Weeks later came Benioff's post and SIForce was born. However, the official documentation and Trailhead still say AIforce, and many pages still show Headless 360.
This mess has practical consequences. If your team searches for documentation, signs contracts or writes runbooks, you will be living with all three names at the same time for quite a while.
Why "SI" and not "AI"
The context matters. Shortly before the announcement, the White House ordered executive branch agencies to replace "Artificial Intelligence" with "Super Intelligence" in their non-statutory communications.
That order does not bind private companies. Salesforce has not published an official explanation linking the rename to the order either, although outlets such as Business Insider, The Verge and Salesforce Ben pointed out the timing coincidence.
What we can say with confidence is that the name does not change the technology. Not a single line of the technical documentation shows a jump in capability from going from AI to SI.
What SIForce (or AIforce) is, in plain words
AIforce is not Agentforce under a different name, and it is not a single product that you buy and install.
Imagine your CRM is a house with every room locked. For years, the only way to use it was to come in through the front door: the Salesforce interface.
AIforce is a live interface layer that opens windows in that house. It lets data, permissions, workflows and agents reach other places: Claude, Slack, custom applications or third-party agents.
The documentation itself says it plainly: AIforce is not a product or a single tool. It is a platform strategy enabled by the Salesforce Headless Toolkit.
The four layers worth keeping in mind
It helps to think of the architecture as a stack of four levels.
Data 360: the context, with structured, unstructured and federated data.
Customer 360: the applications, processes and business semantics.
Agentforce: the agentic workforce that reasons and executes actions.
AIforce / SIForce: the layer that takes everything above to any external interface.
Underneath it all runs the trust layer: OAuth, object and field permissions, sharing rules and the Trust Layer. That foundation is what makes it possible to leave the interface without losing security control.
At Vantegrate we have been building on this platform since before it had a name. You can check our Salesforce page to learn about the full approach.
The components that are already available
The AIforce launch came with three commercial experiences and a toolbox. Let's go through them one by one.
Claudeforce (Salesforce in Claude)
It is a packaged plugin that connects Claude to an org through a secure MCP server. At launch, 37 sales skills ready to use were announced, along with more than 40 skills for developers in the Claude Code plugin.
An important detail for those evaluating budget: Sales Cloud in Claude is still in Beta and requires Claude Enterprise. It is available for all editions, but that requirement is not minor.
Slackforce
It brings CRM context inside Slack. It includes Slackforce Surfaces, Slackbot, Slack CRM and Slack Code, which lets you query records, edit them and build experiences without leaving the chat.
For a sales team that lives in Slack, this means fewer screen switches. Availability of each sub-product should be verified separately.
Agentforce Coworker
It is an AI teammate inside Lightning that reasons over accounts, activity and history. It can execute actions and invoke Agentforce agents that are already deployed.
Salesforce presented it as something existing customers can activate directly, keeping their permissions and business rules. That continuity is precisely the central argument of the whole architecture.
The Headless Toolkit: where everything is built
The toolkit is the set of mechanisms that makes AIforce possible. It is worth understanding even if it gets technical.
Headless 360 MCP Server: an MCP server hosted by Salesforce, in Beta, that requires API v67 or higher and OAuth.
Salesforce DX MCP: more than 60 tools for metadata, Apex, tests, DevOps and code analysis.
Headless Experience Layer (HXL): defines an interface once and renders it in Agentforce, ChatGPT, Claude or Slackbot; it is still in Beta.
Salesforce Multi-Framework: it is already GA and lets you run applications built with conventional web frameworks on the platform.
Agentforce Vibes: the assisted development environment that comes with the DX MCP preconfigured.
One detail that sums up the debate: the MCP server endpoint still contains headless-360. The marketing name changed twice, but the technical endpoint did not move.
For technical teams that want to go deeper, we put together a dedicated section, Salesforce for developers, with the patterns we use in production.
AIforce vs SIForce: the honest comparison
When someone asks for a "before and after" table, the right answer is an uncomfortable one. There is no official specification of SIForce with a column of differences.
What we can do is go through each dimension and say whether a change was published or not.
Positioning: rename, no new technical definition.
Functional capabilities: no additional capability announced.
Integrations: Claude, Slack, Lightning, ChatGPT, Cursor and MCP clients; same list.
APIs and SDKs: there is no "SIForce API" and no version of its own.
AI models: the layer remains agnostic (OpenAI, Anthropic, Gemini, Llama, Mistral and open source).
Privacy and data: same controls, no new specific policy.
Deployment: no new hosting option and no on-premise edition.
Pricing: no new SKU or price book under the SIForce name.
Migration: no migration announced because of the name change.
There is a subtle but fundamental distinction between "there is no change" and "there is no published change." Salesforce could use the new brand to introduce packaging or features later, but today it has not documented that.
That is why at Vantegrate we follow a simple rule: we treat SIForce as the new commercial name for AIforce until a Release Note, a SKU or an API version says otherwise. It is the most defensible position.
The change that does matter: Agentic Identity and HPIs
If the rename is the curtain, this is what happens behind the stage. And it is far more relevant to your budget than any logo.
Salesforce announced that it wants to give every external agent its own identity, separate from the person it helps. It is called Agentic Identity, and it lets you assign the agent a narrower set of permissions than a human user has.
That same identity serves another purpose: attributing traffic and billing for it. Every successful call from a registered agent, through MCP or direct API, is classified as a Headless Platform Interaction (HPI) and shows up in the Digital Wallet.
How much an HPI costs
Here we have to be transparent: it is not known yet. The official Rate Card lists the HPI, but its multiplier is marked as pending announcement (TBA).
What is known is the reference price of Flex Credits: US$500 per 100,000 credits. A standard Agentforce action currently consumes 20 credits, but that does not mean an HPI will cost the same.
Salesforce clarified that agentic metering is not active yet and that it will give at least 30 days of notice before it starts deducting credits. It also confirmed that sandboxes, scratch orgs and Developer Edition will not consume HPIs under this mechanism.
What changes depending on the type of customer
The contractual impact is not the same for everyone.
New customers (who bought after the announcement): they must register their agents within three months of Agentic Identity becoming available.
Existing customers with agents through API: they will register those agents and move to the new billing model at renewal.
Any user of a Salesforce MCP server: mandatory registration within three months of the notice.
Traditional integrations: they keep their current pricing and security terms.
This is key for those who operate in regulated sectors such as financial services, where every identity with access to data has to be audited and justified.
What nobody tells you about privacy
There are two subtleties in the documentation that affect any risk assessment.
The first is the difference between Zero Data Retention and data residency. When you use a model outside the trust boundary, the data does leave temporarily to reach the model; the protection is that the provider does not retain it or use it for training, not that it never processes it.
The second is more specific: the documentation says that pattern-based and field-based data masking is disabled for agents. The argument is that hiding information can remove the context the agent needs to reason.
For companies that handle personal, financial or health information, this has to be part of the threat model. At Vantegrate we explain how we approach it on our pages about platform security and data transparency.
Koa: the model that is actually new
While the name of the layer was changing, Salesforce launched something that is a real technical novelty: Koa, its first reasoning model specific to CRM, built on NVIDIA Nemotron.
Koa is an option within Agentforce, not "the SIForce model." You can choose it at the org, agent or subagent level, and its weights and inference stay inside Salesforce's trust boundary.
For workloads that need the context to never leave that boundary, an internally hosted model has different properties from an external one. It is an architecture decision, not a marketing one.
A ten-step action plan
You have the diagnosis. Now for the specifics.
Do not plan a data migration because of the SIForce name. Your org, Data 360, agents, Flows and Apex stay the same.
Inventory all external agents that access the platform and classify them: MCP, direct API or traditional integration.
Identify every MCP server enabled in Setup and which permission sets have access.
Prepare the org for Flex Credits and Digital Wallet, but do not estimate a cost per HPI until the multiplier is published.
Monitor the formal availability notice for Agentic Identity.
Register each agent with its own identity and the principle of least privilege as soon as it applies.
Rotate the OAuth credentials of MCP clients and API agents so they use the new identity.
Validate everything in a sandbox before production, especially permission failures and human approval paths.
Model costs per process, not per prompt: a single interaction can generate HPI, Agentforce actions, inference and Data 360 consumption.
Keep all three names as aliases in your internal documentation until the endpoints and contracts change.
To understand step 9, think of an agent like Sellium, which manages sales over WhatsApp. Each conversation can query a catalog, create an opportunity and trigger a workflow: three different interactions with three potentially different units of consumption.
The same applies to a conversational BI agent like Metrix, which runs several queries per question, or to Arconte, which processes documents and writes records in batches. The economics of the agent have to be designed before you scale, not after.
The real story is not the name
If we had to sum it all up in one sentence, it would be this: the important news is not AIforce → SIForce, but interface-centered Salesforce → Salesforce accessible by agents.
Headless 360 introduced the architectural change and AIforce gave it a platform brand. SIForce, for now, gives it a name aligned with the trendy terminology.
The relevant technology is still the same: decoupling the CRM from a fixed screen through MCP, APIs, skills and generative experiences, with the platform's permissions and context working behind the scenes. Analysts such as Constellation Research already anticipate that other SaaS vendors will replicate the pattern.
For those building agents, the indicator to watch is not the logo. It is the Release Note that formalizes SIForce, the SKUs, the actual availability of Agentic Identity and the HPI multiplier in the Rate Card.
At Vantegrate we follow each of those milestones closely, because our suite of AI agents runs on exactly this layer. If you want to see what a registered agent looks like, with its own identity and modeled costs, you can schedule a demo with our team.
And if you prefer to keep reading before talking to anyone, on Pulse we will publish every update on this transition as Salesforce documents it. You can learn who we are and why we got into this on our team page.
Frequently asked questions
Does SIForce replace Agentforce?
No. Agentforce is the platform that builds and runs agents; SIForce (formerly AIforce) is the layer that takes those agents, along with data and processes, to external interfaces.
Do I have to migrate anything because it is now called SIForce?
No migration has been announced because of the name change. Existing integrations are preserved and the architecture is still that of the Headless Toolkit.
So what will change?
Agentic Identity and billing by Headless Platform Interactions. That means registering agents, rotating OAuth credentials and enabling Flex Credits in the org, with advance notice of at least 30 days.
How much will each agent interaction cost me?
It has not been published yet. The HPI multiplier is listed as pending in the official Rate Card, so any dollar estimate would be a guess.
Does the change to "Super Intelligence" mean the product is more powerful?
There is no evidence of that. The technical documentation describes exactly the same agentic, headless layer that was announced as AIforce.
Do I need Claude Enterprise to use Claudeforce?
Yes, at least for Sales Cloud in Claude, which is also still in Beta. You should include that requirement in any cost evaluation.
