Rippling this week launched AI Spend Console, a tool that maps AI token spend to individual employees, teams, and roles — after the HR software company discovered it was on track to burn 40% of its R&D headcount budget on AI tokens alone. Spending was compounding at 80% month-over-month, and if unchecked, Rippling would have spent 90% of what it pays its R&D staff on inference tokens the following year. The product ships as an add-on for Rippling HR subscribers and as a standalone integration for other HR systems.
The reckoning came in March, when CFO Adam Swiecicki walked the executive team through the numbers. Chief Product Officer Matt MacInnis described the meeting to TechCrunch as an urgent turning point. An internal audit found that 10-15% of employees were driving 60% of total AI spend, and one engineer was expensing $50,000 a month on tokens.
Rippling's first move was to negotiate spending caps with Cursor, OpenAI, and Anthropic. The immediate problem was behavioral: employees defaulted to the newest, most expensive frontier models for every task, from production code to grammar fixes. There was no routing logic and no cost signal at the point of use.
Key facts
- 01Rippling was on track to spend 40% of its R&D headcount budget on AI tokens, with usage growing 80% month-over-month.
- 02An internal audit found 10-15% of employees drove 60% of total AI spend, with one engineer burning $50,000 a month.
- 03After building an internal AI gateway, Rippling cut token spend to 15% of headcount budget while maintaining ~600 billion tokens of monthly usage.
- 04July's token spend cost only 37% of April's, despite similar volume, after routing prompts to cheaper models like Z.ai's GLM 5.2.
- 05CEO Parker Conrad said GLM 5.2 came in 85% cheaper than frontier models with nearly identical performance in Rippling's internal benchmarks.
MacInnis was blunt about why the tooling gap exists.
The company built its own AI gateway to route prompts to the most cost-effective model for each task, and wrapped it in a dashboard scoring prompts per day against work output — lines of code, pull requests, and peer review quality. The blog post pitching AI Spend Console specifically calls out the ability to flag "engineers with high AI spend whose peers frequently ask them to redo work in code reviews." It is, in effect, an AI slop detector attached to a finance tool.
The routing shift produced the numbers Rippling now uses to sell the product. The company hit a peak of 605 billion tokens in April. In July, internal usage was roughly 600 billion tokens again — nearly identical volume — but the cost came in at 37% of April's bill. MacInnis attributed the drop entirely to routing, quipping that the sales team is no longer allowed to run grammar edits through Fable-tier models.
Founder and CEO Parker Conrad has been public about which models Rippling actually uses. In internal benchmarks, he said Grok was the all-around leader, but Z.ai's GLM 5.2 came in 85% cheaper with nearly identical performance. GLM 5.2 has become a favorite among tech companies for coding tasks, and Databricks has publicly championed it as well. SpaceX's acquisition of Cursor means many of these models are now accessible through a single vendor relationship.
The bigger organizational shift is that Rippling stopped treating AI access as a utility. Engineers, who have been the earliest and heaviest users, are now measured on token spend per unit of shipped work. The company is extending the same model to customer onboarding teams, where productivity is measured in customers onboarded rather than pull requests. Employees identified as effective AI users have been designated "AI captains" and tasked with coaching the rest of the company.
MacInnis was explicit that the productivity link is the gating factor for broader rollout.
The caveat is that this is a vendor pitching its own product with its own numbers. Rippling has not published independent audits of the productivity gains, and the 37%-of-April figure measures cost, not output quality. Enterprises considering AI Spend Console would still need to trust Rippling's gateway with sensitive prompt data, and the tool's spending-control features require using that gateway rather than an existing one. Whether the dashboard actually distinguishes productive AI use from expensive-looking activity is the question every buyer will ask.
The Rippling story is a preview of where enterprise AI budgets are heading. The tokenmaxxing phase — hand every employee a frontier-model seat and hope for productivity — is ending, replaced by gateways, model routing, and per-employee ROI dashboards. That is bad news for frontier labs whose pricing assumes customers cannot see the meter, and good news for cheaper open-weight models like GLM 5.2 that win on cost-per-task. If Rippling's framing catches on, AI access inside companies will stop looking like Slack and start looking like a corporate credit card — approved, monitored, and revoked when the spend does not tie to output.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
