Rippling’s Chief Product Officer Matt MacInnis still remembers the March executive meeting where CFO Adam Swiecicki presented a number that shocked the room: the HR software company was on track to burn 40% of its R&D headcount budget on AI tokens. That meant it was spending as much on AI inference as it paid 40% of its engineering unit’s employees — millions of dollars, with spending growing 80% month-over-month.
“We were incredulous,” MacInnis told TechCrunch. That moment of sticker shock became the origin story for AI Spend Console, a product Rippling unveiled this week to help enterprises track, contain, and justify their AI spending.
Also read: Google warns of vishing attacks targeting US financial firms, extorting millions in Bitcoin
From tokenmaxxing to cost control
Rippling was far from alone in its predicament. The start of 2026 saw a wave of enterprises “tokenmaxxing” — aggressively pushing AI adoption across their workforces without guardrails on cost. Rippling’s internal analysis found that roughly 10–15% of its employees drove about 60% of total AI spend, with one engineer alone spending $50,000 a month on tokens.
The company’s response was not to cut AI usage but to rein it in. Management launched an urgent project to understand where the money was going and what it was getting in return. The first fix was negotiating maximum spending caps with each AI provider — Cursor, OpenAI, and Anthropic. But Rippling quickly discovered a structural problem: employees defaulted to the most expensive frontier models for every task, regardless of complexity.
Also read: Ex-Spotify engineers raise $10M to bring real-time AI personalization to e-commerce
MacInnis was blunt about the incentive misalignment: “The inference providers, like Anthropic and OpenAI, have absolutely no incentives to help you control your spend. They have every incentive for it to be a runaway expense.”
Building a smarter AI gateway
By mid-2026, enterprises had learned two hard lessons. First, they need access to multiple models from multiple labs at various price points — including cheaper open-weight options. Rippling CEO Parker Conrad noted last month that his company’s internal benchmarks found SpaceX’s Grok was the all-around leader, but that “GLM 5.2 is 85% cheaper but nearly identical performance” compared to frontier models. Z.ai’s GLM 5.2 has become a favorite for coding tasks across the industry, with Databricks also championing it.
Second, enterprises need an AI gateway that routes each prompt to the best, most cost-effective model for the task. Rippling built its own gateway as part of AI Spend Console. The tool produces dashboards that score attributes like prompts per day combined with work output — lines of code, pull requests — against spend. It also flags problematic patterns, such as which engineers’ work peers frequently ask to be redone in code reviews.
The results were dramatic. Rippling dropped its token spend from 40% of its headcount budget to about 15%, without curtailing usage. In July, internal usage hit 600 billion tokens — nearly the peak of 605 billion tokens the month the CFO issued his warning — yet the cost was 37% of April’s token spend. “That’s just because now we’re routing to the more effective models,” MacInnis said, joking that “we’re not letting the sales team do grammar updates using Fable.”
Beyond engineering: measuring real productivity
Technology alone isn’t the answer, Rippling says. The company identified employees who used AI effectively and designated them as “AI captains” to assist colleagues across the organization. But extending AI beyond engineering remains a work in progress, MacInnis acknowledged, since software engineers have been the primary users so far.
Rippling is now piloting AI for customer onboarding teams to automate mailing data and data-reconciliation tasks, with the dashboard measuring productivity in terms of customers onboarded. “We have to be able to link token consumption in G&A functions and in customer-facing functions back to productivity,” MacInnis said. “If we can’t do that, all bets are off on any of this stuff being available to the broader employee base.”
The broader implication is significant: if Rippling’s experience is a template, tokenmaxxing may have swung so far that employee AI access is no longer treated like Slack or email. Access may now be contingent on demonstrable productivity gains.
AI Spend Console is included for Rippling’s HR customers, though additional AI usage-based costs apply. It can also be purchased as a stand-alone product and integrated with another HR system of record, MacInnis said.
This article discusses enterprise AI cost management strategies and does not constitute financial advice. AI technology markets and vendor pricing are volatile and subject to rapid change.
