For nearly two years, Salesforce has had one AI message – Agentforce. Every keynote, earnings call, and product launch has told you that autonomous agents will deliver and define the company’s future.
Yet despite Salesforce reporting strong adoption numbers, the conversation across the ecosystem hasn’t kept pace. Everyone is still asking about ROI, product positioning, and real-world deployment or success stories. Where does Agentforce actually sit in a normal business’ AI strategy? Interestingly, a recent study from KeyBlanc found that Salesforce customers do not view Salesforce’s AI plan favorably at the moment, citing that their CIO survey “delivered another blow with Salesforce being a standout for all the wrong reasons”.
That’s what makes Salesforce’s growing relationship with Anthropic so interesting at the moment. While Agentforce remains the company’s flagship AI product in marketing terms, some of its most compelling AI announcements this year have centered on Claude instead. Since expanding their partnership throughout the year, Salesforce and Anthropic have introduced Claude into Agentforce, Salesforce extensions for Claude, Claude Cowork capabilities, and, most recently, Claude Tag, bringing Claude directly into Slack as a teammate.
All of these new features collectively point to something quite big emerging for the CRM giant. If Claude becomes the AI “environment” employees naturally choose to work in, does it also become the most important product in Salesforce’s AI strategy today, even if Salesforce doesn’t own it?
Claude as the Front Door to Salesforce AI
When the two parties first started collaborating, Claude was the AI model sitting quietly underneath Salesforce’s AI ambitions, helping power Agentforce and working behind the scenes. Now, that’s starting to change quite drastically.
Over the past year, Anthropic has steadily turned Claude into something much bigger than a chatbot. Alongside its reasoning capabilities, Claude can now write and review code, work across multiple files, tackle longer-running projects, and collaborate with users through tools like the aforementioned Claude Code and Claude Cowork.
More recently, Claude Tag brought it directly into Slack, allowing teams to bring Claude into conversations with a simple @ mention. These all may seem incremental, but altogether, it shows Claude becoming an interface where work begins, happens, and finishes, rather than just simply generating responses in the background.
And it’s fair to say that Salesforce has capitalized on this shift. Earlier this year, the company expanded its partnership with Anthropic to give Claude access to Salesforce data and allow it to trigger actions inside Salesforce. Instead of asking customers to switch into a desiccated Salesforce AI experience, the company is meeting them where they already work.
A big part of making this possible is the Model Context Protocol (MCP). While the name sounds technical, it simply allows AI assistants like Claude to discover approved business tools, retrieve information from them, and carry out actions without every integration needing to be built from scratch.
Of course, connecting an AI model to your business systems is only half of the battle – a “human in the loop” is still required to decide what data the model is allowed to access and what actions it can perform. Salesforce believes that’s where its long-term value lies.
This also helps explain the thinking behind Headless 360. Rather than expecting customers to work through Salesforce’s own interface, the company is building a platform that can sit behind whichever AI assistant or application people prefer to use. The AI assistant of the hour is, of course, Claude. But could it be another in a few years time? Either way, Salesforce’s flexibility is the real strategy bonus for them.
It’s never suggested its future is dependent on Anthropic, and it already supports multiple AI models across Agentforce. Claude simply happens to be the clearest example (so far) of what Headless 360 looks like in practice.
Salesforce’s enthusiasm for Anthropic also appears to extend well beyond their current product integrations. Following reports from TickerTrends that Anthropic is now tracking at a reported $74.1B annualized revenue run rate (compared to OpenAI’s $41.3B), Salesforce CEO Marc Benioff publicly praised the company on X, calling its enterprise offering a “hat trick” of “the best model (Claude), the best coding agents (Claude Code), & the best productivity tool (Cowork).
He also reiterated that Salesforce is proud to be an early investor, customer, and partner.
Whether the revenue estimates shown are exactly right isn’t really the point. What’s more interesting is that Benioff chose this moment to publicly position Anthropic as the enterprise AI company to beat, as Salesforce continues to deepen Claude’s role across its AI strategy.
Could Claude Save Agentforce, or Expose Its Weakness?
As said, Salesforce would almost certainly reject the idea that Claude is replacing Agentforce, and they’d point to their recent financial success to back that.
In its latest earnings, Salesforce reported that Agentforce has reached $1.2B in annual recurring revenue (ARR), up 205% year-over-year, while Agentforce and Data 360 together now represent a $3.4B ARR business. The company also said customers have generated more than 3.8 billion Agentic Work Units across Agentforce and Slack.
They’re certainly commendable numbers, but the problem is that they don’t really answer the question a lot of our readers are still asking – what exactly is Agentforce?
Over the past year, Salesforce has described Agentforce as a jack of all trades – an AI platform, an orchestration layer, a digital labor platform, a coworker, and a way to build autonomous agents.
These descriptions aren’t necessarily contradictory, but many are finding it hard to understand, especially as Salesforce continues to introduce new AI products alongside it.
Claude, in comparison, feels a lot simpler to many. Open the app, ask it to do something, and let it work. Whether it’s writing code, summarizing documents, researching a customer, or collaborating inside Slack, the experience feels instantly clear and familiar. That’s part of its appeal, and perhaps why Salesforce has become a lot more willing to bring it closer to its platform.
And if we look at it that way… Claude may be solving one of Agentforce’s biggest problems. Rather than expecting customers to interact directly with Agentforce, Salesforce enables Claude to become the interface while Agentforce quietly coordinates everything happening behind the scenes.
Claude handles the conversation, while Agentforce grounds responses in Salesforce data, applies permissions, and ensures actions are carried out securely.
Suddenly, Agentforce becomes less like ChatGPT for Salesforce and more like the operating system sitting underneath whichever AI assistant you prefer to use.
While this may make a lot of sense, it doesn’t come without risk. If employees continue to associate productivity gains with Claude over Agentforce, does it start to face a difficult balancing act? Salesforce desperately needs customers to see enough value in its orchestration and data layers in order for them to continue investing in the platform, even if the AI assistant belongs to someone else.
On top of that, a recent report from CIO stated that many customers are hesitant to adopt Headless 360, driven by unpredictable consumption costs.
“CIOs have become highly sensitive to unpredictable consumption pricing after a decade of cloud cost overruns,” Dion Hinchcliffe, CIO practice lead at The Futurum Group, said. “Enterprises understand the strategic power of headless CRM and agentic workflows, but they are also seeing the possibility of runaway machine-generated activity inside core systems of record.”
While the vision looks amazingly strong, they’ll still need to convince customers that it’s worth paying for.
Final Thoughts
One thing Salesforce has always proven in history is its ability to adapt, never afraid to change direction when the market demands it – and Headless 360 is the latest example of that mindset.
Rather than bettering everything on one AI model or assistant, Salesforce is building a platform designed to work with whichever tool its customers ultimately choose.
Claude is the in-thing right now, but tomorrow, it could be something else. If Salesforce gets that strategy right, it may not matter who owns the AI interface, as long as Salesforce remains the platform powering it.
