The Pricing Experiment That Froze Its Own Customers
Cycling through four distinct pricing models — seat, action, outcome, and compute metering — while a product is in active rollout is not an iteration strategy, it is a symptom of unresolved product identity . Salesforce's 150,000-plus customer base has watched this play out in real time, and the response has been to wait. The gap between the installed base and actual Agentforce adoption is not a marketing problem or a sales capacity problem — it is a signal that customers doing their own cost modeling cannot produce a reliable number, so they are not committing. When cost uncertainty is the top-cited barrier to adoption for an AI product from the world's largest CRM vendor, the product's commercial architecture is the problem, not the market.
The Dumb-Pipe Threat Underneath the Revenue Number
The $1.2 billion AI revenue figure Salesforce announced is the number that gets the press release — but the structural question it cannot answer is whether that revenue reflects durable platform value or LLM pass-through fees that OpenAI or Anthropic could disintermediate at will . One practitioner analysis of the enterprise AI governance race named the threat precisely: Salesforce, ServiceNow, and Microsoft all face the same nightmare of becoming "dumb pipes" while the model providers own the control plane . Agentforce is Salesforce's answer, but a product that hasn't stabilized its own pricing model cannot credibly claim to own the control plane. The revenue figure and the adoption figure tell opposite stories, and the adoption figure is the one that matters for where this ends up.
Layoffs Alongside an AI Revenue Milestone
The optics of announcing $1.2 billion in AI revenue and then cutting staff connected to that product in the same reporting window are difficult to manage . In communities where Salesforce practitioners gather, this sequence was not read as efficient allocation — it was read as evidence that the AI revenue is not producing the organizational confidence the number implies. A company genuinely winning on AI does not trim the teams attached to its AI narrative while simultaneously citing that narrative to investors. The timing produced exactly the kind of interpretive friction Salesforce cannot afford when it is already asking its install base to accept pricing uncertainty on a foundational product.
Slack as Both Asset and Exposure
Slackbot's positioning as "the front door to the Agentic Enterprise" is the clearest version of Salesforce's agentic thesis — and the one with the most competitive pressure bearing on it. The April overhaul added 30 new AI features to Slackbot in its most ambitious update since the acquisition , but coherence is not protection. Microsoft's seven-model and agent sandbox launch this week puts Microsoft's workplace AI directly against Slack in the same enterprise accounts. SAP is offering free design-time access to enterprise AI agents through year-end as an explicit install-base lock-in play. Salesforce bought Qualified to absorb Piper, its AI SDR product , which fills a specific gap — but each acquisition-as-capability-gap signals to practitioners that Agentforce is assembled rather than architected. The Slack bet is real, but it requires Salesforce to win a feature war against a competitor with deeper model relationships and a larger infrastructure budget.
Where the 8,000-Customer Number Points
The adoption rate against Salesforce's installed base is the most honest data point in this story, and it points toward a specific outcome rather than an open question. Customers who are delaying Agentforce adoption are not waiting for a pricing model clarification — they are evaluating whether Microsoft Copilot or SAP's agent layer solves the same problem with less uncertainty baked in. Every quarter that Salesforce's pricing architecture remains unsettled is a quarter where its own install base shops alternatives that already have cost predictability inside the budget line. Salesforce has the customer relationships and the data assets to win this — but it will win by resolving the product identity question, not by adding more pricing models to an already unstable set.