The Tokenmaxxing Hangover: Rippling's New Tool Exposes the AI ROI Void

For the better part of 2026, the corporate mandate has been 'tokenmaxxing'—an aggressive, unchecked push to integrate generative AI into every conceivable workflow. But as the bill comes due, the industry is discovering that without a mechanism to measure actual output, AI spend is less of an investment and more of a leak.

Nowhere is this more evident than at Rippling, as first reported by TechCrunch. The HR software provider recently unveiled the AI Spend Console, a product designed specifically to help companies track and contain AI expenditures. While marketed as a management tool, the product's existence is a tacit admission that the 'AI productivity' promise has, in many cases, failed to materialize as a positive return on investment (ROI).

**The Cost of Unchecked Ambition**

According to reporting from TechCrunch, Rippling's pivot toward cost-containment was born from a moment of genuine executive shock. In March, CFO Adam Swiecicki presented data showing that Rippling was on track to spend as much on AI tokens as it did on 40% of its entire R&D headcount budget.

Chief Product Officer Matt MacInnis told TechCrunch that the executive team was "incredulous" to find that token spending was growing by 80% month-over-month. Had the trend continued, the company projected it would spend nearly as much on tokens—90%—as it spent on the compensation for its high-paid R&D unit employees within a year.

This wasn't a systemic failure of the technology, but a failure of governance. Rippling's internal analysis revealed a stark disparity in usage: roughly 10% to 15% of employees were responsible for approximately 60% of the total AI spend. In one extreme case, a single engineer was spending $50,000 per month on tokens.

**The 'AI Slop' Problem**

Follow the money, and you find a productivity gap. The AI Spend Console isn't just about tracking dollars; it's about identifying "AI slop." The tool maps spending against actual work output, allowing management to see if employees are genuinely more productive or simply generating more noise.

In a blog post, Rippling noted that the tool can identify specific engineers who maintain high AI spend but whose peers frequently request they redo work during code reviews. This suggests that the speed of AI generation is often offset by a decrease in quality, creating a cycle of expensive, low-value output.

MacInnis noted that the incentive structures of the providers themselves exacerbate the problem. He told TechCrunch that frontier model providers, specifically OpenAI and Anthropic, have "absolutely no incentives" to help customers control spending, as they benefit from runaway expenses and provide poor usage insights.

**The Architecture of Containment**

To stop the bleed, Rippling implemented a two-pronged strategy: aggressive negotiation and technical routing.

First, the company negotiated maximum spending caps with its primary tool providers, including Anthropic, OpenAI, and Cursor. They discovered that employees were defaulting to the most expensive frontier models for every task, regardless of whether the task required that level of compute.

Second, Rippling built an AI gateway to route prompts to the most cost-effective model for the specific job. Rippling founder and CEO Parker Conrad observed that while SpaceX's Grok led in overall benchmarks for the company's internal use, Z.ai's GLM 5.2—a Chinese model also championed by Databricks—offered nearly identical performance for coding tasks at 85% of the cost.

The results of this shift were immediate. MacInnis shared with TechCrunch that while internal usage remained high—hitting 600 billion tokens in July, similar to the peak in April—the cost of July's spend was only 37% of April's cost. By routing tasks to more effective, cheaper models, Rippling reduced its token spend from 40% of its headcount budget to approximately 15%.

**Opinion: The End of the AI Free-for-All**

*In my view, Rippling's move marks the end of the 'experimental' phase of enterprise AI. For the last year, AI access has been treated like Slack or email—a utility provided to all. But Rippling is signaling a shift toward a gated model. If a company cannot link token consumption in G&A or customer-facing functions back to measurable productivity, the business case for broad employee access vanishes.*

**The Path Forward**

Despite the technical fixes, Rippling admits that human intervention is still required. The company has appointed "AI captains"—employees who use the tools effectively—to mentor others. However, MacInnis told TechCrunch that expanding AI effectiveness beyond the engineering org remains a work in progress.

Currently, Rippling is attempting to apply these productivity metrics to customer onboarding teams to automate data-reconciliation and mailing tasks. The goal is to link token spend directly to the number of customers onboarded. As MacInnis warned, if this link cannot be established, "all bets are off" regarding the availability of these tools to the broader workforce.

For the market, the AI Spend Console is available to Rippling's HR subscribers (with additional usage-based costs) or as a stand-alone product that integrates with other HR systems of record. It serves as a stark reminder that in the enterprise, the most important metric isn't how many tokens you can process, but how many of those tokens actually create value.