HR software company Rippling was on pace to spend an amount equal to 40% of its research and development payroll budget on AI tokens by early 2026, according to a report published this week by TechCrunch. The company unveiled AI Spend Console, a new product designed to help businesses monitor and control their AI expenditures after discovering employees were rapidly burning through millions of dollars on AI services. Chief Product Officer Matt MacInnis recalled a March executive meeting where CFO Adam Swiecicki presented figures showing monthly spending growth of 80%, a trajectory that would have pushed token costs to 90% of R&D employee compensation within a year.
When Rippling analyzed its spending patterns, it found that roughly 10% to 15% of workers accounted for about 60% of total AI expenses, with one engineer alone spending $50,000 monthly. The firm reached a peak of 605 billion tokens in the month following the CFO's warning. By July, the company used 600 billion tokens again, but the expense dropped to just 37% of April's cost. After implementing its new tracking system and routing infrastructure, Rippling reduced its token spending from 40% of headcount budget to approximately 15% while maintaining usage levels.
The AI Spend Console maps individual employee, team, and role-level spending against productivity metrics, promising to identify "which engineers have high AI spend whose peers frequently ask them to redo work in code reviews," according to the company's announcement. MacInnis told TechCrunch that inference providers like Anthropic and OpenAI "have absolutely no incentives to help you control your spend," adding that they don't offer strong usage insights or collaborate with each other. The product includes dashboards scoring attributes such as daily prompts combined with work output and expenditure, and is available to Rippling's HR subscribers with additional usage-based fees or as a standalone offering integrated with other HR systems.
Rippling discovered that workers defaulted to the most recent and expensive frontier models for every task, regardless of complexity. The company addressed this by negotiating maximum spending caps with tools including Cursor, OpenAI, and Anthropic, then building its own AI gateway to route prompts to the most cost-effective model for each job. CEO Parker Conrad noted that internal benchmarks revealed SpaceX's Grok led overall performance, but that "GLM 5.2 is 85% cheaper but [had] nearly identical performance" to top-tier options. The savings came primarily from smarter model routing rather than usage reduction, MacInnis explained, joking that the company stopped "letting the sales team do grammar updates using Fable."
The company appointed high-performing AI users as "AI captains" to assist colleagues, though efforts to expand AI beyond engineering remain a work in progress, with software developers still the primary users. Rippling is testing applications for customer onboarding teams to automate mailing data and reconciliation tasks, measuring success by the number of customers onboarded. MacInnis warned that if productivity gains can't be linked to token consumption in administrative and customer-facing roles, broader employee access may be restricted. The report suggests that unchecked AI spending may lead companies to treat AI tools differently from universal platforms like Slack or email, granting access only where measurable returns justify the cost. The shift from early-2026 tokenmaxxing to strict cost controls signals that enterprises are moving past experimentation toward selective deployment, prioritizing measurement over enthusiasm when capital is at stake.