Walmart's highest-profile AI experiment just lost its architect. Michael Pfaffenberger, the distinguished engineer who built Code Puppy from a handful of users into a company-wide vibe-coding platform, is heading to AI startup Pydantic. The departure comes as the internal tool he built has reshaped how tens of thousands of Walmart employees—from engineers to hourly store associates—write their own software.
The move raises fresh questions about how enterprises sustain homegrown AI initiatives when the people who built them move on. For project managers and team leads watching the trend, there's plenty to unpack.
From 1,000 Users to 75,000: How Code Puppy Spread
A year ago, Code Puppy barely existed outside Pfaffenberger's immediate team. According to Business Insider, the tool now serves close to 75,000 users across Walmart's merchandising, supply chain, and store management operations. That growth didn't happen by accident—it came from solving real problems across the organization. One regional load manager, for instance, built a tool with Code Puppy that helps truckers avoid unnecessary trips by matching loads to destinations more efficiently.
Walmart's CTO highlighted the tool's broad appeal in June, noting it has found fans everywhere from merchandising desks to store floors. The spread beyond engineering teams is what makes it stand out. Walmart said Code Puppy is now used by roughly as many non-engineers as engineers—a rare outcome for internal developer tools, which typically stay confined to technical staff.
Why the Open-Source Bet Matters
Pfaffenberger didn't just build Code Puppy for Walmart. He open-sourced it, and the external version has been downloaded more than 496,000 times, including over 71,000 downloads last month alone. Dozens of contributors outside Walmart have added features and bug fixes to the public codebase. At the shareholders meeting last month, Pfaffenberger and colleague John Choi received an award from CEO John Furner in recognition of the project's impact.
The open-source approach served a dual purpose. It pulled in outside talent to strengthen the tool while Pfaffenberger was still at Walmart, and it created a community that can carry forward without him. The internal team—Ryan Eccles, John Choi, Kaveri Bakshi, Wesley Blakemore, Andrew Tilson, and TJ Webb—will continue supporting the project under Brian Knapp's leadership.
Our Analysis: What This Means for Enterprise AI Adoption
For project managers tracking AI adoption inside large organizations, this story surfaces two tensions that rarely get discussed together.
First, there's the talent problem. Pfaffenberger's move to Pydantic isn't unusual—builders often want to work where they can move fastest. But when an internal tool depends heavily on one person's vision and energy, that departure can stall momentum. Walmart mitigated this by having Pfaffenberger build a team and codify his principles before leaving. That's smart succession planning, and it's something any leader sponsoring an internal AI initiative should push for upfront.
Second, there's the vendor-lock-in issue. Code Puppy wasn't built to rely on a single AI model or provider. It can switch between dozens of models, distribute workloads, and compare results—exactly the kind of flexibility large companies are hunting for as they try to avoid dependence on Anthropic, OpenAI, or Google. Pydantic CTO David Montague said Pfaffenberger's experience building multi-agent AI tools is precisely what his startup needs as enterprises struggle with adoption.
Say you are a project manager trying to pilot an AI tool in your organization. The Code Puppy story tells you two things: get the right person to build it, but make sure you don't build it in a way that only that person can maintain it.
What to Watch Next
Pfaffenberger's departure is confirmed, but the real test comes next. Will Code Puppy keep growing under his successors, or will it lose momentum without its founder driving it? Walmart is in a quiet period ahead of its August 20 earnings report, so don't expect detailed public commentary for a while. Pydantic, meanwhile, is clearly betting that enterprise AI coordination—the unglamorous work of making multiple AI tools play well together—is the next big market. That focus tracks with what Walmart built.
For readers here, the practical question isn't whether to experiment with vibe-coding tools. It's how to structure those experiments so they survive contact with real organizational turnover. Build the team before you build the roadmap. Document the principles. Open-source what you can.
Key takeaways
- Homegrown AI tools can spread fast when they solve real operational problems across the org, not just for engineers
- Open-sourcing internal tools can attract outside contributions and reduce maintenance burden
- Enterprise AI projects need succession plans from day one—don't let a tool depend solely on one person
- Multi-model flexibility is becoming a key differentiator for internal AI platforms as companies avoid vendor lock-in
Frequently asked questions
What is vibe coding?
Vibe coding refers to using AI assistants to write, edit, test, and manage software through natural-language instructions rather than manual coding. Walmart's Code Puppy is an example that lets both engineers and non-technical employees build their own tools this way.
How many people use Code Puppy at Walmart?
Code Puppy grew from roughly 1,000 users to nearly 75,000 users in about a year, spanning merchandising, supply chain, store management, and hourly associates.
Who is Michael Pfaffenberger?
He was a distinguished engineer in Walmart's Global Tech group and the creator of Code Puppy. He's now leaving for AI startup Pydantic, where his experience building multi-agent AI tools is expected to be central to the company's enterprise strategy.
Will Code Puppy continue after Pfaffenberger's departure?
Walmart confirmed the tool will continue to be supported by his colleagues, including several named engineers and program managers. The internal team is expected to maintain and build on the project.
What makes Code Puppy different from other AI coding tools?
Unlike tools such as Anthropic's Claude Code or OpenAI's Codex, Code Puppy isn't tied to a single AI model or provider. It can work with dozens of models from different suppliers, letting developers switch between them, compare outputs, or distribute workloads across providers.
Inspired by reporting from: businessinsider.com