View / AI is just software

The blast radius from Chinese open-weight model Kimi K3 continues to widen. President Donald Trump’s tech adviser Michael Kratsios and Treasury Secretary Scott Bessent are threatening China over what they deem as IP theft, while OpenAI’s Dean Ball is pushing the White House to stop US companies from using it. On the other side, venture capitalist Bill Gurley has argued “we need to let the free market work.”

Who’s right here? To assess that, it’s helpful to look at the same argument but replace the word “AI” with “software.”

Everyone from White House officials to Silicon Valley executives to, well, my mom, seems to get emotional when talking about AI.

Large language models — especially those trained on other labs’ data — are, in the end, just massive, billion-dollar pieces of software. And innovation in Silicon Valley has often worked best when software itself is free and open source. The money is usually made on the implementation of the software and on everything it enables.

Frontier labs were never going to lock down AI models and charge for access in perpetuity, without ever having to compete. It was inevitable that others would catch up — and even use frontier models themselves to generate valuable data for training competing models. That isn’t theft any more than it was theft when AI labs used my book, for instance, to help train their models. (I am all for it). And Chinese imitation isn’t something that will inherently kill the American frontier labs.

Frontier models, almost by definition, can maintain their lead even if other runners flood the course. Models could begin to recursively self-improve and leap further ahead. The sheer size, rate of improvement, and capability would make AI labs perhaps the most valuable companies in history.

OpenClaw and other agent harnesses, which allowed ordinary people to do incredible things with AI, also showed that the magic isn’t just about the model but everything built around it. And when OpenAI and Anthropic built their own harnesses, demand for tokens exploded. Any model could plug in, but frontier labs had the edge because they tuned their harnesses to their homegrown models.

Once the frontier labs get consumers hooked on, say, Claude Cowork or Codex, their job is to make them easier and easier for customers to use. Then, it’s to build guardrails and infrastructure around them, and make them so good that no user wants to switch to something almost as good, even if it’s a touch cheaper.

And the role of regulators? You can’t stop software at the border, and it’s hard to legally prevent companies from using whatever software they want. Export controls won’t be effective at stopping Chinese firms from distilling US models, even if they’re successful at starving China of inference (a more important goal, anyway, given that’s where most of the economic gains and innovation will come from).

Shutting down access to US software also won’t address the cybersecurity and bioterrorism risks wrought by powerful Chinese AI models. You prevent those risks by hardening your own defenses against those attacks.

In the end, if frontier labs can’t figure out a business model around the most powerful software ever created, they don’t really deserve to survive. (For the record, I think they’ll do fine.) If they falter, investment in massive models will shrink, but innovation won’t end. The greatest minds of our generation will just shift to other breakthroughs. That’s just free markets doing their job. And remember: It’s just software.