Walmart Limits Employee AI Use to Manage Rising Costs

Walmart has begun restricting employee access to its internal AI assistant, Code Puppy, after discovering that costs associated with the large language model (LLM) powering the tool exceeded initial expectations.

The company had previously encouraged its 2.1 million employees to use Code Puppy freely for various workplace tasks, such as spreadsheet analysis and creating presentations. However, Walmart is now allocating a fixed number of AI tokens per employee, restricting how extensively the tool can be used.

A token is a basic unit of computation in language models like those developed by OpenAI and Anthropic; usage of these tokens often correlates directly with overall costs. This policy change comes as the AI industry transitions from flat-fee subscription models to pay-per-use pricing for AI inference.

Walmart has expanded its AI tool offerings and provided training to workers, promoting experimentation with automation to enhance productivity. Despite these investments, the new cost structure is forcing the retailer to address the challenge of measuring and controlling AI-driven productivity gains versus their financial impact.

One issue identified is "token maxxing," where employees maximize the use of AI tools to meet key performance indicators (KPIs), sometimes leading to unnecessary or excessive use. This practice, popularized in some tech circles, increases operational expenses under usage-based AI pricing. Companies have even implemented AI leaderboards to track and reward high utilization, further incentivizing token consumption.

The complexity and resource-intensity of LLMs—especially advanced models capable of recursive problem-solving—can quickly escalate costs. Multi-agent workflows, in which several AI agents collaborate or work iteratively, may also create unpredictable financial burdens due to repeated or inefficient prompting.

Walmart’s updated guidance encourages employees to select appropriate models for specific tasks and to prioritize value creation when deploying AI solutions. The company points out that more sophisticated models should only be used for complex challenges, reserving them from routine tasks that can be handled with simpler, less costly models.

This phenomenon is not unique to Walmart. Other firms such as Uber have already reported reaching their annual AI budgets within a matter of months, while leading AI providers—including Anthropic, OpenAI, and Microsoft (with GitHub Copilot)—have moved enterprise clients to usage-based billing. The shift underscores the need for large companies to align productivity gains from AI with sustainable financial practices.

Walmart’s approach represents a broader trend among major employers striving to balance the benefits of AI-powered automation with the realities of rising operational expenditures and the need for careful workflow management.

Source: artificialintelligence-news.com