Salesforce is introducing monthly AI token budgets based on employees’ roles, after some staff spent thousands of dollars a month on tokens last quarter. In an internal email, Chief Operating Officer Miguel Milano said the policy is intended to direct AI spending toward work with greater value for customers and the business. For companies expanding their use of AI, managing model costs without curbing employees’ output is an increasingly practical operational challenge.
A Salesforce spokesperson said the company has not introduced individual token budgets across the entire workforce and that AI use remains under team leaders’ oversight. That description does not fully align with the role-based allowances set out in the internal email. The spokesperson did not explain how the two approaches fit together.
Technical roles receive up to $1,000 a month
The email sets a monthly budget of $1,000 for technical roles in the sales organization, led by President and Chief Revenue Officer Alexa Vignone. Non-technical sales roles in the organization are allocated $100 a month.
For Salesforce Customer Success, led by Jim Roth, and Salesforce Global Professional Services and Forward Deployed Engineering, led by Lori Steele, technical delivery roles receive $500 a month. Non-technical sales roles in those groups are also allocated $100. Employees are expected to manage their own budgets, while organizational leaders are responsible for monitoring team usage.
Salesforce has already tested the approach with one team. The email said that after a month of real-world use, the team spent less than its budget in the second month without a decline in output. It did not specify the team’s spending, size or the metrics used to assess its work.
Approval requirements rise with spending
As an employee approaches their allowance, the DevBar app will send alerts to the employee and their manager through Slack. Employees can request approval to exceed their individual budget by up to 20%. One-time increases of no more than 20% require approval from a second-line manager.
Requests above 20% of an employee’s budget, but no more than three times that amount, require approval from a third-line manager and a return-on-investment explanation. Approved increases last 90 days and are reviewed when they expire. Requests exceeding three times the individual budget must be approved jointly by Alexa Vignone, Jim Roth or Lori Steele and the finance department, with monthly reviews.
The email also says team leaders can use the team’s overall budget to offset spikes in an individual employee’s usage. Managers are encouraged to review spending monthly and discuss token use in one-on-one meetings. The email also recommends matching models to tasks, with Slackbot and Gemini suggested for drafting, summarization and research. Salesforce said budgets should not prevent high-impact work and that employees can seek exceptions.
In countries and regions with employee committees, managers will be able to see only aggregated team spending, not individual usage data.
Other companies are setting AI spending limits
Coinbase sets weekly AI spending caps according to employees’ roles and seniority, ranging from $500 to $5,000. JPMorgan Chase has set a monthly allowance of $2,000 for some engineers using Anthropic’s Claude Code.
Salesforce Chief Executive Marc Benioff said in May that the company expected to spend about $300 million on Anthropic tokens this year, mainly for coding agents. As AI tools move from trials into day-to-day work, spending controls affect both model costs and employees’ ability to keep using the tools for their tasks. The precise scope of Salesforce’s new allowances remains unclear alongside the spokesperson’s statement that AI use is managed by team leaders.