Walmart's Code Puppy dodges AI vendor lock-in, spreading coding to 75,000 staff

Walmart distinguished engineer Michael Pfaffenberger built Code Puppy, an internal AI coding assistant on the Pydantic AI library that lets anyone from engineers to hourly store associates write, edit, test and manage software with natural-language instructions. A year after it had about 1,000 users, he wrote in a farewell memo reviewed by Business Insider, lifetime user count was approaching 75,000.

Unlike Anthropic's Claude Code or OpenAI's Codex, Code Puppy is tied to no single model: it works with dozens of models from different suppliers — developers can switch between them, compare results, run several at once and spread workloads to dodge usage limits and control costs. CTO Suresh Kumar called it a way technology's power has gotten democratized: merchandising, supply chain and store managers all use it, about as many non-engineers as engineers. One regional load manager built a tool that analyzes hundreds of truckloads to cut drivers' empty returns.

Pfaffenberger called the decision to open-source Code Puppy pivotal: 496,000+ downloads including 71,000+ in the last month, dozens of contributors unaffiliated with Walmart, and heavy use in the Cerebras Discord. He left for AI startup Pydantic in August 2026; a named six-person team keeps the tool running inside Walmart.

Walmart wanted to avoid the defining AI-era problem of getting locked into a handful of powerful providers.

Model-agnostic routing means usage limits, pricing changes and outages at any single provider stop being existential.

Natural-language coding opened tool-building to non-engineers, which is where most of the user growth came from.

When a strategic input is controlled by a few vendors, standardize on your own layer that treats those vendors as interchangeable parts.

Pfaffenberger left for Pydantic in August 2026; Walmart confirmed the departure and its internal team continues to support the project. The open-source version passed roughly half a million downloads. At the annual shareholders meeting, Pfaffenberger and colleague John Choi received an award from CEO John Furner for the project.