According to a TechCrunch report, HR software provider Rippling has unveiled "AI Spend Console," a product designed to combat the phenomenon of "tokenmaxxing" and help companies track and contain their artificial intelligence costs. One of the most compelling features of this new tool is its ability to map exactly how much individual employees, teams, and specific roles are spending on AI tools, while evaluating whether they are genuinely becoming more productive or simply generating low-quality AI output, often referred to as "AI slop." In its blog post, Rippling promises that the new tool will highlight, for instance, "which engineers have high AI spend whose peers frequently ask them to redo work in code reviews."
The background of the tool's development stems from Rippling's own decision earlier this year to go all-in on "tokenmaxxing"—as did many other organizations—only to realize very quickly that its employees were burning cash at a staggering rate.
The Data That Shocked Rippling’s Management
Rippling's Chief Product Officer, Matt MacInnis, recalled a March executive team meeting in a conversation with TechCrunch. During the meeting, CFO Adam Swiecicki presented a figure that left attendees dumbfounded. According to Swiecicki's data, Rippling was on a fast track to burn approximately 40% of its research and development (R&D) headcount budget on purchasing AI tokens. In practical terms, the company was spending an amount equivalent to 40% of the total compensation and salaries paid to employees within that unit—a sum totaling millions of dollars. As is widely known, the R&D department is home to software engineers at most technology firms.
These expenses were growing at a rate of 80% month-over-month. Had this trend continued at a similar pace, the company was projected to spend nearly the same amount on AI tokens in the following year as it did on the salaries of its expensive development staff—amounting to roughly 90% of the department's headcount budget. "We were incredulous," MacInnis told TechCrunch. Management immediately launched an "urgent" project aimed at thoroughly understanding the nature of these expenditures and evaluating exactly what the company was getting in return for such massive sums of money. As part of the marketing campaign for the new product, the launch video features CFO Swiecicki sitting on a stool while employees collect stacks of cash and dump them directly into a paper shredder.
Analysis of Usage Patterns and AI Provider Behavior
When Rippling conducted an internal analysis of its usage patterns, it uncovered several surprising facts. In a blog post, the company shared that "roughly 10% to 15% of our employees were driving about 60% of total AI spend. One software engineer alone hit $50,000 in a single month." Rippling clarified that it did not want to halt employee usage of AI tools entirely, but rather to rein it in significantly.
The company's first step was to negotiate maximum spending caps with each of the tool providers its employees were using, namely Cursor, OpenAI, and Anthropic. During this evaluation, the company immediately identified a glaring issue: employees tended to default to the newest and most expensive "frontier models" for every single task, regardless of its complexity.
MacInnis criticized these model providers, telling TechCrunch: "The truth is that 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, and that’s exactly what they do. They don’t provide you with great usage insight, and they don’t collaborate with one another." According to the report, this issue was highly common in early 2026, but now, eight months into the year, many enterprises have already recognized and adopted several key operating principles.
Practical Solutions in the Market: Multiple Models and Smart Routing
First, enterprises now realize that they need to leverage a variety of models from different AI labs at various price points, including the option to use leading open-weight models, sometimes of Chinese origin. Rippling founder and CEO Parker Conrad noted last month that when his company conducted internal benchmarks for its own use cases, it discovered that SpaceX’s Grok model was the overall leader in performance. However, it also became clear that the "GLM 5.2 model is 85% cheaper but has nearly identical performance" to the leading frontier models. (SpaceX now owns Cursor, a platform that provides access to Grok and dozens of other models.) Z.ai’s GLM 5.2 has recently emerged as a highly favored Chinese model for coding tasks among tech companies, with Databricks also championing and supporting its adoption.
Second, enterprises now understand that they require an "AI gateway" that automatically routes prompts to the best, most cost-effective model for each specific task. Rippling reached this same conclusion and subsequently built its own AI gateway integrated into the newly launched product. MacInnis explains that organizations already using a different AI gateway can still utilize the AI Spend Console product. However, if they wish to access the features that govern and actively manage actual spending, they will need to switch to Rippling’s proprietary gateway.
Reducing Spend and Measuring Productivity
The AI Spend Console produces dashboards—which during the "tokenmaxxing" era were referred to as "leaderboards"—that score attributes based on a combination of metrics, such as the number of prompts per day, actual work output (such as lines of code written or pull requests), and cost.
Thanks to the implementation of this tool within Rippling, the company reported that it successfully reduced its token expenditures from 40% of its development headcount budget to just about 15%. However, the company stresses that it did not curtail or limit the volume of AI usage by its employees. MacInnis shared that the company peaked at a consumption of 605 billion tokens in the month the CFO issued his warning. In July, internal usage reached 600 billion tokens again, yet "the cost of July's token spend was 37% of the cost of April’s token spend." MacInnis explained that "that’s just because now we’re routing to the more effective models," jokingly adding, "we’re not letting the sales team do grammar updates using Fable."
At the same time, Rippling points out that technological solutions alone are not enough to resolve the issue. The company identified employees who were using AI highly effectively and appointed them as "AI captains" tasked with assisting the rest of the organization in utilizing the tools correctly. MacInnis admits that efforts to scale AI usage beyond engineering departments remain a work in progress, as software engineers have been the primary users of these tools so far. Currently, Rippling is working on implementing the system within its customer onboarding teams to automate tasks associated with mailing data and data-reconciliation. The dashboard will then measure the productivity of this usage based on a higher volume of newly onboarded customers.
"We have to be able to link token consumption in G&A functions and in customer-facing functions back to productivity. If we can’t do that, all bets are off on any of this stuff being available to the broader employee base," MacInnis says. Therefore, if we extrapolate from Rippling's experience to the broader market, the "tokenmaxxing" phenomenon may have swung so sharply in the opposite direction that employee access to AI will no longer be taken for granted like access to email or Slack. If companies cannot measure an employee's productivity when using the tool, access may not be granted to everyone.
As for the product itself, the AI Spend Console is included for subscribers of Rippling's HR services, though additional costs exist based on actual AI usage. MacInnis adds that the tool can also be purchased as a standalone product and integrated with other human resources management systems.