OpenAI’s next fight shifts from models to lock-in

AI’s next – OpenAI and Anthropic are increasingly competing on how hard it is for customers to leave—moving beyond model performance to “sticky” AI work platforms. A Pydantic CEO says the economics are changing as profit margins matter more and frontier labs can’t keep ra

For more than a few years, the AI industry’s headline has been model quality—who can build the smarter system, faster, and with fewer compromises. But in boardrooms where big customers sign contracts, the argument is starting to change.

Samuel Colvin. CEO of AI startup Pydantic. says the next battle is less about raw model performance and more about customer lock-in. “A year ago, what they cared about was revenue,” Colvin told MISRYOUM recently. “Now, when one assumes they’re both trying to IPO, their profit margin becomes really important.”.

That shift in what matters—revenues versus profit margins—changes the math of competition. Competing only on model quality is expensive, Colvin argues. Frontier labs must spend billions training ever-better models that are soon emulated. “That’s not a great recipe for durable profits,” he said.

So the pressure moves toward something different: products that are harder to leave even after the novelty of a chatbot wears off. Colvin believes OpenAI and Anthropic are increasingly focusing on building tools that customers can’t easily swap out. “They are doing their very best to find ways of locking people in that are not related to model quality. ” he said. “That’s where I think Claude Code and Codex and all that work is coming from.”.

Coding services, in this view, are not just a feature—they’re a business model. Tools like Claude Code and Codex can be used in ways that look more like ongoing software work than short conversations. Developers can “quickly consume huge numbers of tokens. ” Colvin said. while running Claude Code or Codex on complex projects—generating far more usage. and revenue. than a typical chat session.

The stickiness gets even stronger once code starts to multiply. Claude Code and Codex can help companies generate software at unprecedented speed. creating codebases that grow beyond what human developers can realistically manage. The risk for customers is that they may become dependent on the same AI tools to maintain. update. and understand the software those tools helped create.

That help explain why Anthropic and OpenAI are racing to turn Claude Code and Codex into broader AI-powered work platforms rather than standalone coding products. Anthropic is pushing in this direction with Cowork, while OpenAI is planning to merge Codex into ChatGPT.

But while AI makers chase lock-in, enterprise buyers are often moving in the opposite direction.

Walmart provides a clear example. Its home-grown coding assistant. Code Puppy. is designed to avoid dependence on any single AI provider and give the retail giant more control over its codebase. The system can switch between models from OpenAI, Anthropic, Google, and others, helping Walmart control costs and reduce vendor lock-in.

That contradiction is increasingly shaping the AI market. OpenAI and Anthropic want sticky products with high margins that keep customers inside their ecosystems. Enterprise customers want flexibility, portability, and lower token bills.

The next winners may be the companies that navigate that conflict best—offering enough value to be useful day after day, without making it impossible for customers to change direction.

OpenAI Anthropic Samuel Colvin Pydantic Claude Code Codex Claude Code and Codex Cowork ChatGPT Walmart Code Puppy AI coding assistants customer lock-in vendor lock-in IPO tokens enterprise AI