Salesforce held its yearly Dreamforce conference this week, and the message from CEO Marc Benioff was that AI agents are adding to human work, not replacing it. He unveiled new agents with names like Albert, Piper, and Casey, pushed Slack as a place where people and AI agents build software together, and highlighted Claudeforce, the expanded partnership with Anthropic that puts Claude's AI inside Salesforce tools.
Benioff told the audience that customers are not cutting jobs because of agents. Instead, he said, the hours saved are handed back to employees. That is a comforting message for a room full of corporate technology leaders worried about what AI means for their teams.
But it does not match what Benioff has said elsewhere about his own company. In a podcast interview earlier this year, he said Salesforce shrank its customer support staff from 9,000 people to 5,000, cutting about 4,000 jobs as AI agents took over the work. He put it plainly: he needs fewer people now. So the company is telling customers one thing about job losses while doing the opposite inside its own walls.
This gap is not just embarrassing, it is useful information. It suggests that when a company gets serious about using AI agents for repetitive support work, headcount does fall, whatever the sales pitch says on stage. Any business leader watching Salesforce's actions rather than its keynote should expect the same pressure on their own support, service, and back office teams over time.
The other loose thread is pricing. Salesforce describes Claudeforce as billed per user, per month, which sounds simple and predictable. But the actual structure being rolled out looks different: businesses pay Salesforce for how much they use its systems through automated calls, and pay Anthropic separately for the AI's actual thinking, with no cap on that second bill. Salesforce's own team has said there is no single number you can currently ask for.
That is a real problem for anyone running a budget. Traditional software pricing was easy to plan for: a fixed number of seats, a fixed monthly cost. Usage-based AI pricing means the bill can move depending on how much work the AI agents do, and nobody controlling the budget can predict that in advance until the invoices start arriving.
This shift is not unique to Salesforce. Across the software industry, companies that used to charge by the seat are quietly moving toward charging by usage, because AI agents mean fewer human seats are needed but more automated actions happen behind the scenes. For any business that relies on enterprise software, the seat count on your invoice may soon matter less than the volume of automated work you generate.
The bigger economic backdrop makes this worth watching closely. Corporate profit margins in the United States recently hit a record share of the economy, while the portion going to workers as pay fell to its lowest level since records began in 1947. Productivity gains from technology like AI tend to protect company profits first, not paychecks. Dreamforce's agent demos were the easy part. The pricing model and the headcount question are the parts that will actually show up on a CIO's desk next quarter.