Salesforce and Anthropic have taken another step in integrating artificial intelligence into enterprise software with Claudeforce, a partnership that will enable Claude to query up-to-date Salesforce data, reason about business opportunities, and perform certain actions within the CRM. The initial feature will be Salesforce in Claude, a plugin with 37 preconfigured capabilities for sales, currently in pilot phase with an open beta scheduled for September 2026.
The essentials of Claudeforce and AI in CRM in 20 seconds
- Salesforce in Claude launches with 37 commercial capabilities, such as reviewing opportunities, preparing meetings, or analyzing the pipeline.
- Claude will be able to act on Salesforce respecting permissions and business rules.
- The integration uses MCP, API, and the AIforce infrastructure.
- Microsoft Dynamics 365 and HubSpot are also bringing AI agents directly into their CRMs.
- The trend is moving from the assistant that responds to the agent that queries data and executes tasks.
This news fits into a transformation already affecting much of the Customer Relationship Management (CRM) market. Generative AI initially arrived at these platforms to draft emails, summarize meetings, or retrieve information via natural language. By 2026, the next move is turning those assistants into agents connected directly with data and business processes.
Salesforce isn’t alone. Microsoft Dynamics 365 is developing a similar architecture around Copilot and its agents, while HubSpot allows access to CRM data from Microsoft 365 Copilot and subsequent execution of actions within HubSpot.
The technological race in CRM is shifting from who stores commercial data best to which platform enables AI to understand and work with that data without losing control over permissions and business rules.
Claude moves from querying Salesforce to acting on it
The first implementation of Claudeforce is called Salesforce in Claude. Anthropic and Salesforce initially prepared 37 capabilities focused on the sales cycle.
These include preparing meetings, reviewing the health of an opportunity, or analyzing the pipeline. Claude can also use available context on accounts and opportunities to generate user-specific interfaces.
The most technically interesting part occurs when moving from querying information to modifying it.
Salesforce plans that actions requested from Claude are executed through the platform itself, ensuring that the company’s rules and permissions continue to apply. The model can interpret a request and decide which tool it needs, but this doesn’t mean bypassing the existing authorization framework.
This separation will be crucial for deploying agents in enterprise environments. A model that responds incorrectly to a question is problematic; one with permissions to modify opportunities, clients, forecasts, or processes could cause much greater issues.
Management of the integration also changes. According to companies, an administrator can connect Salesforce with Claude centrally and manage authentication and permissions for all users.
The underlying infrastructure is called AIforce. Salesforce envisions it as a layer providing data and processes to different agents via Model Context Protocol (MCP), APIs, and command-line tools.
Salesforce is also expanding Headless 360 to turn app capabilities into resources discoverable and usable by authorized agents through open standards.
This is more significant than simply adding another chatbot to the CRM sidebar: the enterprise application begins exposing its functions so that autonomous software can utilize them.
Microsoft and HubSpot advance in the same direction
This trend is clearly visible among other major providers.
Microsoft already describes Dynamics 365 as a platform combining agents, Copilot, and integrated AI functions in CRM and ERP. Its initial wave of 2026 updates will extend these capabilities to sales, customer service, and Customer Insights.
In Dynamics 365 Sales, Microsoft leverages data from both the CRM and Microsoft Graph. The salesperson can work in natural language with business information, emails, and meetings, as well as receive recommendations and alerts about pending actions.
Microsoft 365 Copilot’s Sales Agent goes even further because it works from applications like Outlook and Teams. It can query account and opportunity information, draft communications with CRM data, draw conclusions from meetings, and update specific CRM fields without requiring the salesperson to switch applications. It also supports both Dynamics 365 Sales and Salesforce.
HubSpot follows a similar approach. Its HubSpot Agent for Microsoft 365 Copilot allows natural language queries of CRM data and uses that information to generate analyses before taking actions in HubSpot. For example, a sales team can review open deals and sort them by expected close date.
| Platform | Integrated/Connected AI | Use Cases |
|---|---|---|
| Salesforce | Claude, Agentforce | Pipeline, meetings, opportunities, and CRM actions |
| Microsoft Dynamics 365 | Copilot and agents | Sales, accounts, meetings, service, and CRM updates |
| HubSpot | Proprietary AI and connection with Copilot | Record querying, analysis, prioritization, and actions |
| Salesforce + Claude | Claudeforce | Initial 37 capabilities and governed execution of actions |
Implementation and availability vary, so the table doesn’t imply functional equivalence. But it highlights a common direction: AI is progressively approaching the transactional layer of CRM.
From email-writing copilots to agents using tools
During the early phase of enterprise generative AI, many functions centered around text generation: drafting a sales email, summarizing a conversation, or preparing a customer response.
The second phase added access to corporate context. The model stopped working solely with prompt input and began querying documents, emails, and knowledge bases.
The current evolution introduces a third element: actions.
An agent can interpret a request, consult multiple sources, reason about the results, and leverage tools to complete part of the process. In a CRM, this might mean retrieving an opportunity, reviewing recent interactions, summarizing its status, and then updating information or initiating another authorized flow.
MCP is gaining importance especially at this layer. The protocol aims to provide a standardized way to connect models and agents with tools and external sources, reducing the need to build completely new integrations for each combination.
The adoption of AI in CRMs also raises less visible issues beyond model performance. Identity, permissions, traceability, data quality, and governance of actions become part of the AI challenge.
CRMs contain especially sensitive information: clients, contracts, forecasts, conversations, prices, opportunities, and potentially personal data. Granting access to an agent requires defining not only what it can query but also what it can modify and under what circumstances human intervention is needed.
Claudeforce is noteworthy because it reflects this transition. Salesforce provides the data, permissions, processes, and rules; Claude supplies the reasoning layer. The approach is to combine both without making the model the ultimate authority over the system.
Claude will also have a reverse presence. Salesforce uses it within Agentforce and has made it a key component of Slack. Furthermore, the company will make Claude Code and Claude Enterprise available to developers and knowledge workers as preferred tools.
Salesforce in Claude is not yet a feature available to all customers. It is currently used by pilot clients, with an open beta expected in September. Additional preconfigured capabilities are expected to begin rolling out by late 2026.
The race in AI-powered CRM is thus just entering a new phase. The question is no longer just which platform has the most advanced model but also how many processes that model can support, what data it can work with, and what controls exist as it moves from recommending actions to executing them.
Frequently Asked Questions
What is Claudeforce?
Claudeforce is the extension of the Salesforce and Anthropic partnership to integrate Claude with Salesforce’s data, processes, and actions. Its initial implementation includes a plugin with 37 capabilities focused on sales.
Can Claude modify Salesforce data?
The integration allows actions to be performed on Salesforce, but these are governed by permissions and rules of the platform. It does not grant Claude unrestricted access to the CRM.
Are other CRMs also incorporating AI agents?
Yes. Microsoft is integrating Copilot and agents into Dynamics 365, while HubSpot offers proprietary AI functions and allows CRM data queries from Microsoft 365 Copilot.
When will Salesforce in Claude be available?
It is currently available for selected pilot clients. Salesforce and Anthropic plan to launch the beta in September 2026 and add new preconfigured capabilities starting late this year.
Source: Noticias inteligencia artificial
