A Company Divided Against Itself
Salesforce built its empire on CRM software – the unglamorous but indispensable plumbing of enterprise sales, service, and marketing teams. For two decades, that model printed money. Now the company is steering hard into AI agents, a category it calls Agentforce, and the pivot is starting to cost something real. Not just in dollars, but in the coherence of a product story that used to be remarkably simple to sell.
Agentforce is Salesforce’s bet that autonomous AI agents – software that can complete multi-step tasks without human instruction – will replace or augment the human workflows that CRM software was designed to support. The logic sounds elegant in a keynote. In practice, it puts Salesforce in the uncomfortable position of asking customers to pay for technology that competes with the licensed seat counts driving its existing revenue model.
The Seat Model Is Under Pressure
Traditional enterprise software pricing is built on users. More employees using the platform means more licenses sold, and for Salesforce that arithmetic has worked reliably across economic cycles. Agentforce disrupts that math directly. If an AI agent handles what a team of five service reps previously managed, the rational enterprise customer starts asking whether they need as many seats – and Salesforce has no clean answer to that question without cannibalizing its own revenue base.
Salesforce has tried to reframe this tension by positioning Agentforce as additive rather than substitutive. The company argues that agents expand what teams can do rather than replace those teams outright. But that framing only holds as long as customers accept it, and enterprise procurement teams are not famous for accepting vendor logic that happens to benefit the vendor. The moment buyers start running their own numbers, the seat model faces a real reckoning.
Salesforce’s fiscal results have begun to reflect the strain. Revenue growth has slowed compared to earlier cycles, and management has pointed to deals being restructured around AI consumption models rather than traditional subscription tiers. Consumption-based pricing sounds modern and flexible, but it introduces revenue unpredictability that Wall Street has historically punished in enterprise software. The company is trading a reliable annuity for a variable income stream on the promise that AI volume will eventually compensate for lower per-seat billings.
Agentforce’s Technical Ambition Versus Enterprise Reality
Agentforce is technically impressive on its own terms. It integrates with Salesforce’s Data Cloud to give agents access to unified customer data, and the architecture is designed to let agents act across Sales Cloud, Service Cloud, and Marketing Cloud without manual handoffs. For organizations already deeply embedded in the Salesforce ecosystem, this is a genuine capability advance.
The gap between demo and deployment, however, has been a recurring theme in enterprise AI broadly. Agents that perform well in controlled environments frequently struggle when exposed to the messy, inconsistent data conditions of real enterprise operations. Salesforce is not uniquely vulnerable to this problem, but it is the company standing at the front of the room making the boldest promises – which means it bears the largest reputational risk when deployments take longer or deliver less than anticipated.
A Legacy Business That Still Pays the Bills
The irony in Salesforce’s position is that its legacy CRM business remains fundamentally healthy. Service Cloud and Sales Cloud still dominate their respective categories. Thousands of enterprise customers run critical revenue operations on Salesforce infrastructure, and they are not leaving. The switching costs are prohibitive, the integrations are deep, and the institutional knowledge embedded in years of Salesforce configuration is not something companies walk away from lightly.
That stickiness is exactly what makes the Agentforce pivot so strategically complicated. Salesforce does not need to win AI to survive – it needs to win AI without losing the stable, high-margin business that funds everything else. Every dollar of R&D redirected toward Agentforce is a dollar not spent maintaining and improving the core platform that existing customers actually depend on. And existing customers notice when a vendor’s attention shifts.
There is also a competitive dimension that Salesforce cannot afford to underweight. Microsoft’s Copilot integrations are running directly inside the enterprise tools millions of workers already use daily. HubSpot is pressing into mid-market accounts with AI features at price points well below Salesforce’s floor. ServiceNow is building agent capabilities into IT workflows with the same “we already own the ecosystem” logic Salesforce uses. None of these competitors are ceding ground while Salesforce figures out its internal pricing contradictions.
Salesforce CEO Marc Benioff has been unambiguous about the company’s commitment to Agentforce, describing it in public forums as the most important product in the company’s history. That kind of declarative confidence tends to close off strategic retreat as an option – which means Salesforce is now locked into making this bet work regardless of the short-term pressure it places on sales cycles, deal structures, and the legacy revenue lines that still account for the vast majority of what the company earns. The real question is not whether Agentforce can eventually generate meaningful revenue. It is whether Salesforce can keep its existing customer base satisfied and its seat-count revenue from eroding faster than AI consumption revenue grows to replace it.
The company’s next two or three earnings calls will reveal a great deal about how enterprises are actually adopting agents versus how they are describing adoption in vendor conversations. Those two things are rarely the same number.
