AINEWS
The HR software company's CFO walked into a March meeting with a number that shocked leadership. Its CPO says the AI labs behind the models have 'absolutely no incentives' to help fix it.
In a March executive meeting, Rippling's CFO Adam Swiecicki told the leadership team a number that stopped the room. The HR and payroll software company was on track to burn 40% of its R&D headcount budget on AI tokens. Those are the per-chunk fees AI labs charge each time a model reads or writes a piece of text. The next-year projection pointed to 90% of R&D payroll, according to reporting from TechCrunch.
The company had been an aggressive AI buyer. Its internal analysis found that roughly 10% to 15% of employees were driving about 60% of total AI spend. One engineer was spending $50,000 a month. Token costs were growing 80% month-over-month; if the trend held, the projected AI bill would have equaled nearly all the compensation Rippling paid to its engineering unit. Every number in this story is company-disclosed. None has been independently benchmarked against peers.
Rippling's response is a product it calls AI Spend Console. The tool maps AI spending by individual employee, team, and role. It surfaces which high-AI-spend engineers have peers frequently asking them to redo work in code reviews, a peer-redo signal meant to distinguish spending that pays off from spending that does not. The launch ad leans into the alarm: a CFO sits on a stool while employees shred wads of cash. The product itself is careful to distinguish what it measures (spend plus a code-review signal) from what it implies (productivity).
CPO Matt MacInnis told TechCrunch that inference providers, the companies that actually run the AI models, like Anthropic and OpenAI have "absolutely no incentives" to help customers control spend, and "every incentive" for runaway expense. The providers, he said, do not give customers usage insight and do not collaborate on cost control. That is a structural claim from a buyer who is also selling a tool that responds to the problem, so it lands as labeled commentary, not neutral fact.
Rippling did not wait for the labs. The company negotiated max-spending caps with several major AI providers and shifted engineers off the most recent, most expensive frontier models by default. Internal analysis had found employees defaulting to those frontier models for tasks where cheaper or smaller models would have done the job, a pattern the company's own staff described as "tokenmaxxing," with the wasted output dismissed as "slop." The negotiated caps, not any vendor dashboard, are what set the ceiling.
The case study is real but bounded. The $50,000-a-month engineer may be a unicorn. MacInnis is the executive selling the very tool the story is about. And the asymmetry Rippling points to, high-spend engineers with low peer-redo signals versus high-spend engineers with high peer-redo signals, is the only one of these claims that the product itself can adjudicate. Spend is a different ledger from productivity, and the company knows it.
What to watch: whether any frontier lab ships a customer-facing usage dashboard that lets buyers see per-team, per-model spend without building their own. Until then, the gap between AI productivity and AI bills stays with the customer to instrument. MacInnis's claim, that the labs are content to let it stay that way, is the part worth carrying past this one case.