Rippling unveils AI Spend Console after burning millions on tokens

Rippling has launched a new product designed to help companies control runaway AI bills after the HR software maker said its own internal usage was consuming millions of dollars in tokens within months. The new AI Spend Console tracks who is using AI, how much they are spending, and whether that activity is actually improving output.

The product matters because it reflects a growing enterprise reality in 2026: many companies embraced AI quickly, then discovered that enthusiasm for frontier models can create a severe cost problem unless usage is measured, routed, and tied to productivity.

Rippling says the tool was built after the company itself found that a relatively small slice of employees was responsible for a disproportionate share of AI consumption. The company then cut its token bill sharply without reducing overall usage, largely by directing work to cheaper models that were still good enough for the task.

What Rippling announced

Rippling’s AI Spend Console is a dashboard and controls layer aimed at enterprise AI budgeting. It gives managers visibility into consumption across employees, teams, and roles, while also trying to connect AI spend to actual work output.

In practical terms, the system is meant to answer a question many companies are now asking: are employees using AI to move faster, or are they just generating expensive digital clutter?

The company says the product can highlight patterns such as engineers with high AI spend whose code is repeatedly sent back for revision in code review. That kind of comparison, Rippling argues, can help companies identify high-value usage and curb waste at the same time.

Why this product exists now

Rippling says the launch grew out of a painful internal lesson. Like many technology companies, it embraced heavy AI use early in the year and then realized the spending was escalating far faster than expected.

Chief Product Officer Matt MacInnis said the turning point came in March, when CFO Adam Swiecicki showed executives a projection that stunned the leadership team. The company was on pace to spend about 40% of its research-and-development headcount budget on AI tokens, meaning AI inference costs were heading toward the equivalent of nearly half of the compensation bill for one of the company’s most expensive organizations.

That was only the beginning of the problem. Rippling’s AI bill was rising about 80% month over month. If that pace continued, the company estimated that token spending could approach 90% of its R&D payroll costs over the following year.

MacInnis said the team’s reaction was disbelief, and the company immediately started an urgent effort to figure out where the money was going and what it was buying in return.

How bad did Rippling’s AI spending get?

Rippling’s own numbers show how quickly token consumption can spiral when access is broad and guardrails are weak. The company said roughly 10% to 15% of employees were responsible for about 60% of total AI spend. In one case, a single engineer was reportedly spending $50,000 a month on tokens.

At its peak, Rippling used 605 billion tokens in the month when leadership first flagged the issue. By July, internal usage had again reached around 600 billion tokens, but the cost was far lower than before because the company had changed how that traffic was routed.

That distinction is central to the story. Rippling did not shut down AI use. Instead, it tried to make AI use economically rational.

How did Rippling cut its bill without cutting usage?

Rippling says it reduced token spending from about 40% of its headcount budget to around 15% by combining purchasing controls, model routing, and internal management changes.

The company started by negotiating caps with the major tools its teams were using, including Cursor, OpenAI, and Anthropic. It then discovered a common pattern: employees were defaulting to the newest and most powerful models for every task, even when smaller or cheaper models would have been adequate.

To address that, Rippling built its own AI gateway, which serves as the routing layer inside AI Spend Console. The gateway can direct prompts to different models depending on the task and the cost-performance tradeoff.

MacInnis said companies can still use another gateway if they already have one, but Rippling’s spend controls work best when its own routing layer is used. In other words, the product is not just a reporting dashboard; it is also part of the decision path that determines where requests go and what they cost.

Why AI gateways matter for enterprises

Enterprise AI use is increasingly less about choosing one model and more about orchestrating several. That means companies want a mix of frontier systems, lower-cost alternatives, and in some cases open-weight models that can be deployed more flexibly.

Rippling says it reached the same conclusion that many other businesses have reached in 2026: the best model for a task is often not the most expensive one.

That shift is already visible across the industry. Some enterprises now benchmark multiple models internally and route tasks based on price and performance. Rippling founder and CEO Parker Conrad recently said the company had found Grok to be the strongest overall performer in its testing, but also noted that another model, GLM 5.2, was about 85% cheaper while delivering nearly equivalent results in some evaluations.

The broader point is that cost optimization has become a first-class AI strategy. For companies with thousands of workers, even small per-query savings can add up fast.

What the dashboards measure

AI Spend Console produces dashboards that go beyond raw cost totals. Rippling says the system scores usage patterns such as prompts per day, output volume, and spending, and can compare those inputs with signs of productive work.

The company previously described these internal scoreboards as leaderboards during the era of “tokenmaxxing,” when many organizations encouraged heavy AI usage as a badge of progress. The new framing is different. It is no longer about maximizing usage for its own sake; it is about proving that usage is worth it.

That approach could prove especially important for teams outside engineering, where productivity is harder to quantify than lines of code or pull requests.

Who gets the most value from AI so far?

Rippling says software engineers have been the biggest beneficiaries so far, which is consistent with how AI has spread across the enterprise software world. Coding assistants are easy to deploy, easy to measure, and often produce visible time savings.

Other functions, however, are more complicated. The company is experimenting with customer onboarding workflows, where AI can help automate mailing data and reconcile records. In that case, the relevant output metric is not code volume but how many customers can be onboarded more efficiently.

MacInnis said the company needs a credible way to connect token consumption in general and administrative roles, as well as customer-facing functions, back to productivity, warning that broader access becomes hard to justify if that connection cannot be made.

That is a significant statement because it suggests that enterprise AI access may not become as universal as email or Slack. Instead, organizations may limit access based on measurable business returns.

Why enterprise AI budgets are changing

Rippling’s experience illustrates a broader recalibration underway in corporate AI adoption. In the first wave of enthusiasm, many companies treated AI access as something employees should simply have, then worried about usage later. Now, more firms are realizing that unrestricted access can create a cost structure that is difficult to defend to finance leaders.

The shift is also changing the way enterprises buy AI. Companies are no longer just evaluating model quality; they are comparing model quality, routing layers, controls, and total cost of ownership. That means the market is moving from simple chatbot adoption to infrastructure for managing AI consumption at scale.

In that environment, products like AI Spend Console are not only about saving money. They are about giving finance, engineering, and operations teams a shared language for deciding where AI should be used, by whom, and at what price.

What Rippling is signaling to customers

By commercializing a tool built out of its own internal cost crisis, Rippling is also sending a message to prospective customers: AI can be productive, but only if it is governed like any other major spend category.

The company says the product is included for customers with Rippling HR subscriptions, though additional usage-based charges apply. It can also be sold separately and integrated with another system of record, which could make it attractive to companies that want the control layer without adopting Rippling’s full HR suite.

That positioning matters because the demand is likely to come from companies that have already experienced AI budget shock and now want a better way to manage it.

How Rippling changed its own internal culture

Technical controls were only part of the solution. Rippling says it also identified employees who were using AI effectively and elevated them into informal internal champions, or “AI captains,” to help coworkers learn better practices.

This is an important detail because it shows the company views AI governance as partly cultural, not just mechanical. Better usage depends on better habits, and better habits often spread through peer influence rather than policy memos.

At the same time, the company appears to be narrowing the space where unconstrained experimentation is acceptable. That could be interpreted as a warning to employees everywhere: AI access may survive only where it can be justified through measurable output.

What Rippling’s case says about the AI market

Rippling’s story is useful because it captures both sides of the 2026 AI adoption cycle. First comes excitement, then broad usage, then a bill that forces discipline. In Rippling’s case, the response was not to abandon AI but to treat it as an operational system that needs controls, routing, metrics, and governance.

That is likely to resonate with other enterprises facing similar pressure from finance teams. As AI tokens become a line item with real budget impact, companies are likely to demand dashboards that can answer a simple question: what did we get for the money?

For vendors, that means the next battle is not only about model quality. It is about proving cost efficiency, finding the right model for the right task, and building management layers that keep AI spend from becoming unbounded.

Rippling’s own journey from rapid token burn to internal controls to a commercial product suggests that enterprise AI budgets are entering a more disciplined phase. The company’s message is blunt: if you cannot measure value, you cannot justify unlimited access.

Timeline of Rippling’s AI spend shift

The company’s internal response unfolded quickly over the first seven months of 2026, moving from surprise to cost controls to a product launch.

• Start of 2026: Rippling broadly embraced AI tools and usage rose rapidly.
• March 2026: CFO Adam Swiecicki presented a budget warning that shocked leadership.
• Spring 2026: Rippling negotiated spend caps and began routing work to cheaper models.
• Mid-2026: The company built an AI gateway and added dashboards to track spend versus output.
• August 2026: Rippling introduced AI Spend Console as a product for customers.

Bottom line

Rippling’s AI Spend Console is both a product launch and a warning from inside the enterprise AI boom. The company says it discovered that heavy model use can quickly consume budgets, but also showed that strong governance and smarter routing can preserve productivity while cutting costs dramatically.

For companies now facing the same problem, the message is clear: the next phase of AI adoption will be defined less by how much employees use it and more by whether leaders can prove that the usage pays for itself.