Claudeforce: What Salesforce's Biggest Bet Says About the Future of Software

On Wednesday we showed Salesforce down 31% for the year, run over by the thesis that AI commoditizes software. Today we tell its answer, announced in late August under a name that hides no ambition: Claudeforce. In one sentence: the company that invented cloud CRM made Claude, Anthropic's model, the default reasoning engine of its entire ecosystem. It is no longer an AI feature on the menu; it is AI becoming the core of the product. (Transparency: Claude is also the model used in producing this blog; the analysis below stands on the announced facts, each checkable at the source.)

What was announced, piece by piece

• Claude inside Salesforce. Claude becomes the reasoning model of Atlas, the brain of Agentforce (Salesforce's agent platform), powering Agentforce Coworker and Vibes by default, running inside Salesforce's own trust boundary via Amazon Bedrock. Translation: the agents that serve, sell and operate inside the CRM now think with a frontier model.
• Salesforce inside Claude. The reverse path too: a plugin connects Claude to the live CRM, with 37 prebuilt sales skills: meeting preparation, deal health review, pipeline analysis, governed record updates. If you read our article on skills, this is exactly that concept, packaged by two giants and sold as product: the how-to-sell manual, made executable.
• Claude in Slack. Claude becomes Slack's default model (Slackbot and assistants), gluing AI to the place where teams already talk. Salesforce cites 8.1 million annualized hours of productivity gained with Slackbot, more than double the previous quarter.
• Money and timeline. According to coverage of the deal, about US$ 300 million in Claude token consumption in 2026, on top of an equity stake in Anthropic valued at roughly US$ 5 billion. Pilots running now, open beta expected in September, more skills through year-end.

Why Salesforce did it

Put it next to Wednesday and the logic closes itself. The market is saying the system (the screen, the database, the form) became a commodity and value migrated to reasoning. Salesforce had two exits: spend years trying to build its own frontier model, chasing rivals with years of head start, or embed the best available reasoning and defend what only it has: decades of customer data, distribution and corporate trust. It chose the second, and paid up to choose early. As one analysis of the deal put it, this is also the end of "model-agnostic" enterprise AI: picking one engine and integrating deep beat the "we work with any model" pitch.

For Anthropic, the mirror image: a frontier model alone does not reach the end user either; a distributor with millions of corporate seats is the shortest path to real work. Each side delivered what the other lacked. It is the classic marriage of commodity and distribution, and it signals the consolidation likely to mark the next two years: every major platform choosing its engine.

What changes for users (and competitors)

• If your company uses Salesforce: Agentforce agents and Slack get visibly more capable without you switching systems; it is worth testing the beta on a real sales process and measuring. And Wednesday's contract question applies: with agents doing people's work, how many seats will you renew?
• If you sell software: the message is the whole week's: a reasoning engine is no longer a differentiator, it is an input. The differentiator became the data, the workflow and the trust around it. Partnerships like this will repeat in every vertical.
• If you work in sales or operations: the 37 prebuilt skills are a catalog of what the market considers automatable in your day. Reading that list is reading the map of what stops being human work, and what remains: relationships, negotiation and judgment, as always.

The risk exists and is symmetric: Salesforce ties its product to one vendor's roadmap, and corporate clients come to depend on a model they do not control. But the announcement confirms the thesis this blog has been assembling week by week: value is migrating from the system to the reasoning, and from generic reasoning to reasoning with context. Whoever has the data fits the engine; whoever has the engine goes looking for data. The rest is commodity.