Force Majeure

The New Narrative

A few months ago, Salesforce introduced a new narrative around Agentforce: agentic AI is a boardroom conversation now, not an IT one, with budget migrating from IT to the business, data staying where it already lives thanks to "Zero Copy," and Salesforce itself going headless.

Seen from a little further back, it reads a bit differently: the repositioning of a company that invented enterprise cloud CRM, now adjusting to a shift in where software's value actually sits—a shift it didn't start. Going around IT to reach the business is a familiar Salesforce play, and this time it also reflects how much harder IT has become to sell this particular pitch to. Going headless follows the workflow that was already leaving the application, as much as it sets a new vision for where it should live. "Zero Copy" lets Salesforce query data sitting in an external warehouse like Snowflake without importing it—a sign that most of a customer's data had already moved outside Salesforce's own database, as much as a new capability.

Call it Spinforce, while we're at it.

Force

What actually stayed with me was a single word, not the strategy: workforce, Salesforce, Agentforce. "Force" runs through all three, and its meaning slides a little further each time. In workforce, it names a mass of human labor—a unit of productive power, the way horsepower named a unit for engines. Salesforce built a brand on top of that meaning, capturing "sales force" as a name. Agentforce slides the word again, toward agents as the new unit of that force: deployable, scalable, substitutable. Less augmentation than substitution: the word turns out to be conveniently liquid, meaning human labor whenever that flatters the brand, and agent labor the moment agents become the better story to tell.

Claudeforce

That was spring. In late August, Salesforce and Anthropic announced Claudeforce, and the word did something else entirely this time. Salesforce in Claude ships 37 prebuilt skills—meeting prep, deal health, pipeline review—so a seller can work entirely from inside Claude, connected once by an admin, without opening Salesforce again. Claude reasons over live revenue data and takes action; the action still runs through Salesforce's own permissions, business rules, and audit trail. Two layers that used to be fused into one product and one company have just been officially separated: Salesforce keeps the system of record, the permissions, the governance. Claude gets the interface, the reasoning, the orchestration.

In 2024, a seller worked through Salesforce's own interface, into its own workflows, onto its own data. In 2026, the seller works through Claude, which calls Salesforce's skills and data through MCP. Not long after that, an agent could just as easily be calling Salesforce, SAP, ServiceNow, and three internal databases in the same request—and Claude has no reason to prefer any one of them over the others. That's the risk Salesforce is running, and its own leadership seems to understand it: rather than defend the interface, the company appears to be betting on becoming the thing underneath it—fifteen years of CRM data model, permissions, business rules, audit trail, and customer records. The moat moves from "my people use Salesforce" to "my people's agents execute inside Salesforce." Early results support the bet, for now: Agentforce and Data 360 combined revenue is near $3.9 billion annualized, up over 210% year over year, and Salesforce's stock jumped more than 20% on the announcement—then kept climbing, up over a third within the month.

The Orbit

What the market actually priced in looks different up close. Salesforce frames the deal as a smart defensive move, securing the layer underneath before a competitor's interface gets there first. But a pop that keeps climbing for a month reads less like a bet on Salesforce's own moat than a bet on the orbit it had just locked itself into: proof that Salesforce anchored itself to a body with more gravitational pull than its own. Salesforce doesn't disappear in that reading, any more than a moon disappears once it's captured—it just stops being the center of anything. Gravity pulls both ways between two objects, but momentum belongs to whichever one is moving faster—a comet weighs nothing next to a planet and still bends its path. Salesforce isn't watching that orbit from the outside, either: Salesforce Ventures has been quietly building a stake in Anthropic since a $50 million check in 2023—a stake worth about $5 billion today, and most of the earnings beat the market is rewarding came from marking it up. The word gave the whole reading away for free.

Workforce, Salesforce, Agentforce, Claudeforce all borrowed capitalism's word for labor. It turns out to be physics' word for gravity too.

The same week tested that exact question again, at a smaller scale. Bending Spoons agreed to buy Miro—real brand, real installed base, nearly 4 million paying users—for $1.36 billion in enterprise value, roughly 2.3 times its $600 million in annual revenue. That's a 90% discount to the $17.5 billion Miro was worth in 2022. None of that proves AI destroyed the value of installed software—the drop could reflect plenty of other things, from a bloated 2022 valuation to compressed SaaS multiples generally. But it puts an uncomfortable price on the distinction between having millions of users and owning something genuinely difficult to reproduce. Having a big installed base and being hard to rebuild turned out to be two different things, and the market just put a number on the gap. Salesforce got the same question from the other direction four days after its own rally: shares dropped roughly 4% on September 8, when GPT-6 Astra's launch revived the identical fear across every software stock with a workflow to defend. Same week, same question, asked twice—once through an acquirer's cash, once through a sell-off—and the market answered it two different ways within four days of itself.

Unlearning

The part of this that looks genuinely risky sits one level below the strategy. Salesforce is teaching Claude how to use Salesforce. Claude is teaching the user that they no longer need to know Salesforce at all.

Once a request can just be "prep my Thursday meeting, pull the open opportunities, update the forecast with what I just told you," the person asking stops needing to know, and eventually stops caring, whether the answer came from Salesforce, HubSpot, SAP, three homegrown databases, or something else entirely. Swapping the backend stops costing anyone a second thought. The enormous UX investment the SaaS industry has made over twenty years—the screens, the objects, the dashboards, the training programs—turns from a moat into something closer to sunk cost: paid for, but no longer where anyone is looking.

Call it Fadeforce: the one that means slowly dissolving.

Nobody's Force

The word moves again here, somewhere stranger this time, and what actually stopped me is that it did so twice in a few months, not once over a decade. "Force" already went once from a mass of human labor to a substitutable unit of agent labor. This time it lands somewhere that belongs to neither company fully: Claudeforce is a joint name for a joint capability, coined by two organizations that just agreed force no longer lives entirely inside either of their own products.

The name ends up documenting, almost by accident, the exact shift it's announcing—power no longer sitting in one layer, but split across whichever companies can make the handoff between reasoning and record-keeping work well enough that nobody notices the seam. Whatever the next name in that sequence turns out to be, it probably won't take as long to arrive as this one did.

The test this leaves behind sits underneath the question of whether to deploy Claudeforce, and it's smaller and more uncomfortable:

How much of any system like Salesforce could disappear from the screen a person actually looks at, while the system itself stays exactly where it is, doing exactly what it already does, underneath?