Meta's Adam Mosseri thinks companies will soon put spending caps on AI tools, engineer by engineer

Key points

• Adam Mosseri, head of Instagram, predicted in 2025 that companies will manage AI tool spending per engineer the same way they manage salaries.
• AI tokens, the tiny units of text an AI model processes each time someone uses it, cost money every time an engineer runs a query or generates code.
• Mosseri said engineers could soon face hard limits on how much AI they are allowed to use each month.
• The prediction signals that AI tool costs are already large enough to worry senior executives at companies like Meta.

Adam Mosseri, the executive who runs Instagram for Meta, has a prediction that will interest anyone who works in tech: the bill for AI tools is about to get personal.

Speaking publicly, Mosseri said he expects companies will one day track and cap AI token spending on a per-engineer basis. A token, in this context, is the smallest unit of text that an AI model reads or writes. Every time a developer asks an AI coding assistant a question, the system burns through thousands of tokens, and each one costs a fraction of a cent. Those fractions add up fast across an entire engineering team.

The comparison Mosseri reached for is blunt: treat it like payroll. Salaries are the single biggest cost on most company books, and finance teams track them obsessively. Mosseri thinks AI compute costs, meaning the money spent running AI tools day to day, are heading toward the same level of scrutiny.

TechCrunch AI first reported the remarks.

What does this mean for engineers using AI tools right now?

For most developers today, AI coding assistants feel unlimited. You open a chat window, ask for help, and the answer appears. No meter runs visibly in the corner.

That will likely change. If Mosseri is right, companies will start assigning monthly token allowances to individuals, much like a corporate travel budget or an expense card limit. An engineer who burns through their allocation writing code with AI assistance might have to wait, or justify the spend to a manager.

It is not a fringe concern. Companies across Silicon Valley are discovering that equipping hundreds of engineers with AI coding tools can run to millions of dollars a year in API fees, the charges vendors collect each time their model answers a question. For a large organisation, that is a budget line that demands control.

For engineers, the practical implication is to start paying attention to which tasks genuinely need the AI and which are habit. A quick syntax check probably does not need a full reasoning model. A complex architectural review might. Learning to match the tool to the task, before the company assigns someone else to track it for you, is the sensible move.

For anyone outside the tech industry, this is a useful signal about how real the costs of AI have become. The technology is not free to run, and the companies building products on top of it are working out how to keep those costs from ballooning.