Walmart Limits Use of Code Puppy to Control AI Costs

Walmart Limits Use of Code Puppy to Reduce Expenses

Walmart has recently decided to restrict the use of its internal AI assistant, Code Puppy, due to higher-than-expected costs. Initially, employees were encouraged to use the tool without any restrictions or stipulations regarding usage limits. Now, each employee is assigned a fixed number of AI tokens. This measure aims to control expenses related to AI, as large language models (LLMs) increasingly shift to a pay-per-use model.

Code Puppy was introduced as a tool capable of assisting employees with various tasks such as analyzing spreadsheets and creating presentations. However, the intensive use of this assistant has led to unforeseen costs, prompting Walmart to revise its usage policy.

A Cost-Driven Policy Change

With approximately 2.1 million employees, even modest requests can lead to significant costs for Walmart. The pay-per-use model replaces the old fixed-price subscription system, which offered nearly unlimited access to AI tools. This internal policy change reflects a broader trend among large companies seeking to balance costs and productivity gains associated with AI.

Walmart has expanded the use of AI tools within the company and has provided training to its employees on how to use AI, encouraging workers to experiment and adopt successful practices. Walmart's direction is to promote the use of AI where it can create value while providing guidance on selecting the appropriate tools for each task. Employees have access to other company-funded AI platforms.

The Challenges of Tracking AI Productivity

Part of the problem may stem from the methods used to measure productivity in AI-based workflows. Tracking the number and complexity of AI tool usage as a measure of productivity has led many employees to "gamify" their KPIs (key performance indicators). This practice, known as "token maxxing," was encouraged by a partner at Sequoia Capital in a statement to the Wall Street Journal in April of this year. AI leaderboards have emerged within companies to celebrate those who make the best use of AI software, highlighting the growing importance of these practices.

Larger models, which perform recursive actions, consume more tokens, increasing costs for users. Therefore, Walmart encourages its employees to choose models wisely to avoid unnecessary expenses on simple tasks like analyzing spreadsheets or creating presentations.

The Token Billing Model Gains Traction

Multi-agent AI work can also generate unexpected costs for employers. When employees use iterative loops across multiple agents to achieve a result, the cost of suboptimal outcomes can be measurable in financial terms. By limiting token usage per employee, Walmart aims to control its costs, encourage more judicious use of AI tools, and establish clearer return on investment metrics.

Many AI providers, including Anthropic and OpenAI, have already adopted this token billing model. Microsoft has also announced that GitHub Copilot will be billed according to this model starting June 1. This growing trend highlights the financial challenges companies face with AI adoption, as evidenced by Uber, which has already exhausted its AI budget for 2026.