Salesforce moves the selling workflow into Claude as 37 prebuilt skills

A buyer can forecast a seat count a year out. Forecasting agent consumption across an application vendor's price list and a model vendor's price list is a different exercise. Acharya's argument is that per-seat licensing loses coherence once agents rather than people consume the software, and that consumption metering is the successor ^[10]. Where the application vendor and the model vendor are different companies, he wrote, the customer gets two meters and two invoices for one workflow ^[11].

The account is a Forbes Tech Council column by Vivek Acharya, described there as an AI strategist, author and AI ethics assessor ^[21]. It attributes the position that the interface was never the asset to "Salesforce's own leadership" and "its executives" without naming them. Those executives are described as saying that people use the platform more, not less, when they work through an agentic interface ^[4][5]. The column does not report adoption figures for Claudeforce or any change to Salesforce's published pricing ^[22].

For decades, a packaged application was a database wrapped in an opinionated interface. That interface constrained users to legal actions, enforced a shared definition of every metric it displayed and gave two people the same artifact to argue over ^[15]. This is the part of the argument that stands on its own: remove the screen and each of those has to be delivered somewhere else. Acharya's example is a board meeting where 12 executives arrive with 12 self-built dashboards, each technically correct and none reconcilable with the others ^[16].

Salesforce still holds the system of record, and 37 skills inside another company's assistant is a distribution deal ^[2][17]. The first pressure lands somewhere else. Agents call APIs and invoke tools, and Salesforce exposed its platform to any agent as APIs, MCP servers and command-line tools months before the partnership ^[6]. Acharya wrote that a vendor whose functionality cannot be called, through APIs and open standards such as the Model Context Protocol, is invisible to the fastest-growing category of user ^[9].

A renewal this quarter can settle whether the vendor can say whose authority an agent acts under, what it is allowed to write and what the audit trail shows afterwards. Acharya expects that answer to decide which vendors win regulated industries ^[14]. The pricing question runs longer. Oracle and SAP are making versions of the same system-of-record case while it does ^[7][8]. On the two invoices, Acharya wrote: "No vendor has solved that yet. The one that does will have an advantage worth more than any feature." ^[12]