OpenAI and Anthropic Want to Own the Intent Behind Your Code

Samuel Colvin, CEO of Pydantic — the company behind one of the most widely used Python data‑validation frameworks — has a prediction that should make every developer pause. In a ¹ interview published June 5, Colvin laid out what he sees as the next frontier of AI vendor lock‑in: databases of coding intent.

The idea is simple: corporate subscribers to tools like Claude Code or Codex would get automatic storage of every interaction between developer and AI — every prompt, every reasoning trace, every context of what was being attempted. Click any line of code in your codebase, and you'd see the full exchange that produced it. Useful? Absolutely. Exportable? No. And that's the point.

Colvin, whose company works closely with both OpenAI and Anthropic and recently raised $12.5 million led by Sequoia Capital, describes a deliberate strategy shift from the frontier labs. "A year ago, what they cared about was revenue," he told.¹ "Now when one assumes they're both trying to IPO, their profit margin becomes really important."

The current play: aggressive pricing on coding subscriptions at roughly $200 per month, even though the underlying inference costs much more. The goal is to grow market share and accumulate AI‑generated codebases so large that humans cannot maintain them independently. As Colvin told:¹ "Once customers have these enormous code bases, which would be basically written AI, you get to a point where you can't maintain them as a human. If I've used AI to generate 20,000 lines of code overnight, I can use a model to go and fix that, but as a human, I can't ever go and maintain that code."

The intent database takes lock‑in a step further. Instead of just storing your code, it stores the reasoning behind every line. Colvin described it to:¹ "Imagine if I could click on that line of code and see the full exchange that my colleague had with the AI model to write that line of code, along with all of the reasoning."

This is genuinely valuable. A developer debugging a complex codebase could understand not just what a line does but why it was written that way — the original problem, the alternative approaches considered, the tradeoffs. But the database would be proprietary and non‑exportable. Once your entire engineering workflow depends on it, switching tools means losing access to the institutional memory of how your code was built.

Colvin told:¹ "I think this idea of basically 'we store your trajectories and we give you some database of your trajectories' is attractive and valuable. Those two things are not necessarily the same, but on this occasion, it will be actually valuable as well as being attractive."

The $200‑per‑month pricing for tools like Claude Code Max and Codex isn't about making money on subscriptions — it's about building an installed base. A developer who wrote about dropping Claude Code Max after testing Codex noted that using both tools together cost $20-$120 total and produced better results than $200 for one alone, according to a detailed comparison post.

But the real cost for providers is the inference itself. Running large language models to generate thousands of lines of code costs far more than the subscription fee. Colvin told ¹ the labs are "doing their very best to find ways of locking people in that are not related to model quality. That's where I think Claude Code and Codex and all that work is coming from." Competition on model quality alone is a race to spend billions training better models while providing inference as cheaply as possible — an unsustainable margin killer.

No AI lab has publicly confirmed plans for an intent‑database feature. Colvin's forecast is an informed prediction, not a product announcement, as Let's Data Science notes in their analysis. And his company, Pydantic, is model‑agnostic — it benefits when developers avoid single‑vendor lock‑in, so his warnings come with a commercial angle.

Still, the pattern is clear enough to warrant attention. The signals to watch: official announcements of telemetry features, pricing tiers that offer "free" data storage, export controls or proprietary embedding stores that would make migration difficult, and any API that stores developer interactions without providing a portable export format. The lock‑in Colvin describes doesn't happen through contracts or pricing — it happens through architecture. And once your codebase depends on a proprietary reasoning database, switching costs become existential.

Colvin's prediction fits a broader shift in how AI companies compete. A year ago, the game was model quality — who has the smartest LLM. Today, as top models converge in capability, the battleground has moved to the developer experience layer: editor integrations, agent orchestration, memory systems, and now, potentially, the very metadata of how code gets written.

This is not inherently sinister. Intent databases could genuinely improve code quality, onboarding, and debugging. But when a feature is both "attractive and valuable" — as Colvin described it to ¹ — and locked to a single vendor — developers need to decide whether the benefit is worth the dependency. The question for builders isn't whether these tools are useful. It's whether you want the company that stores your code's intent to be the same company that sets the price for accessing it.