The software giant plans role-based caps after some staff were spending thousands of dollars a month on AI tokens
Salesforce is reportedly putting its employees on an AI allowance.
The company plans to roll out monthly AI token budgets across a range of roles starting in November 2026. The aim is to rein in what leadership sees as runaway spending on AI tools.
What the budgets look like
AI models process text in small chunks called tokens, and providers bill by the token. The more an employee prompts, generates, and iterates, the bigger the bill gets.
According to the reporting, the caps will vary by role and by which part of the organization an employee sits in. The breakdown looks like this:
- Sales, technical roles: $1,000 per month
- Sales, non-technical roles: $100 per month
- Customer success and professional services, technical roles: $500 per month
- Customer success and professional services, non-technical roles: $100 per month
The gaps between tiers are wide. A technical seller gets ten times the budget of a non-technical colleague on the same team. In customer success and professional services, the technical-to-non-technical ratio is five to one.
Warnings, wiggle room, and the spending that started it
The system is not designed as a hard wall. Employees will get notifications through the DevBar Slack app as they get close to their monthly limit.
Staff can also request a 20% overage. Run the math and a technical sales employee could stretch to $1,200 in a month, while a non-technical employee tops out at $120.
The trigger for all this reportedly came from the top. Chief Operating Officer Miguel Milano flagged that some employees were spending thousands of dollars a month on tokens in the last quarter.
Those heavy users were concentrated in sales and customer success, according to the reporting. Those are the same groups now getting the most structured limits.
The bigger AI bill
The per-employee caps sit against a much larger number. Salesforce expects to spend approximately $300 million on tokens from Anthropic in 2026, according to the research findings.
Most of that spending is focused on engineering, where AI coding tools have become central to how software gets built. The company has tied much of its internal AI push to productivity gains through its Agentforce platform.
Salesforce has pointed to results from that push before. The company previously reported $100 million in annualized support cost savings from using Agentforce internally.
Engineering, where the bulk of the roughly $300 million in token spend is concentrated, is not the focus of the caps described in the reporting. Sales and services teams are.
What this means
The role-based design means that rather than issuing a single company-wide cap, Salesforce is making a judgment about where AI produces the most return per dollar. Technical roles get more room because their work presumably benefits more from heavy model usage.
There are risks. Employees bumping into limits might ration their AI usage in ways that slow work down. The 20% overage and early warnings are designed to soften that, but the real test comes once the system goes live.
The things to watch are straightforward. First, whether the November 2026 rollout happens on schedule. Second, whether Salesforce reports any change in its token spending trajectory afterward. Third, whether the $100 million savings figure from internal Agentforce use grows, which would help justify the overall AI bill.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

1 day ago
11








English (US) ·