Rippling makes payroll and HR software, and it just turned its own AI spending disaster into a product it now sells to other companies.
Here is what happened inside Rippling first. At the start of this year, the company told every employee to use AI tools with no limits. By March, its finance chief told the executive team that the company was on pace to spend as much on AI subscriptions as 40 percent of the entire budget it pays engineers, designers, and product staff. Spending was growing 80 percent a month. At that rate, next year the AI bill would nearly equal the payroll itself.
When Rippling dug into the numbers, it found the spending was wildly uneven. A small slice of employees, somewhere between 10 and 15 percent of the workforce, accounted for 60 percent of the total bill. One engineer alone ran up $50,000 in a single month. Worse, nobody could say whether any of that spending was making the work better or just faster to produce and slower to fix.
Rippling's fix had two parts. First, it stopped letting employees default to the newest, most expensive AI model for every task, the way most companies still do today. Instead it built a system that automatically sends simple tasks to cheaper models and only sends hard problems to the expensive ones. Second, it linked spending data to actual output, like how much code an employee produced and how often that code had to be redone. That let managers see who was using AI well and who was just generating expensive busywork.
The results are real. Rippling cut its AI bill from 40 percent of that budget down to about 15 percent, while keeping usage flat. It now sells this system, called AI Spend Console, to other companies through its HR platform.
Rippling's story is not an outlier. Uber recently capped every employee at $1,500 a month per AI coding tool after blowing through its entire annual budget for those tools in four months. Microsoft cut off one of its most popular AI coding tools for a large part of its staff after usage pushed costs sharply higher. A pattern is forming across large companies: 2025 was the year of giving everyone AI access with no rules, and 2026 is the year finance departments are pulling that access back and asking for proof it is worth the money.
There is a broader lesson here for any business, not just software companies. Giving staff a tool with an open-ended bill and no way to measure what comes out the other end is the same mistake companies made with unmanaged cloud computing spending a decade ago. Costs balloon quietly until someone in finance asks a simple question nobody can answer.
The bigger shift to watch is what this means for how companies treat AI access itself. For the last two years, giving every employee an AI subscription looked like giving them email or a laptop, a basic tool nobody questions. Rippling's experience suggests that era may be ending. Going forward, AI access is more likely to work like a company car: useful, valuable, but tied to a track record that shows you use it well.