Anthropic’s Jack Clark on AI’s Missing Brake Pedal—and What Comes Next

The Self-Writing Machine

Here’s the detail that should make you stop scrolling.

Anthropic’s Claude—one of the most widely used AI assistants on the planet—is already running on code that is 80% self-generated. Clark says reaching 100% is plausible within two years. At that point, the system writes itself. Entirely.

That’s not science fiction. That’s a product roadmap.

The implications aren’t just technical. When an AI system can improve its own architecture without human input, the feedback loop accelerates in ways that become genuinely hard to audit, govern, or reverse. Clark’s concern isn’t hypothetical—it’s operational.

The Oil Parallel Worth Taking Seriously

Clark didn’t arrive at this conversation empty-handed. He drew a deliberate historical comparison: AI today resembles the oil industry at the turn of the 20th century—powerful, fast-moving, and shaped by the personalities of a handful of barons rather than any coherent public framework.

Society’s answer to oil wasn’t to stop drilling. It was to build regulation that made the benefits accessible without requiring blind trust in whoever happened to be running the companies.

“That’s clearly where we end up here,” Clark said.

It’s a reasonable analogy. It’s also a convenient one for a company that’s simultaneously warning about risk and preparing for what could be one of the most valuable IPOs in stock market history—with a private valuation approaching $1 trillion.

The Policy Gap Nobody’s Rushing to Fill

The tension in Clark’s position is hard to ignore.

Anthropic welcomed Trump’s recent executive order on AI—a document notably light on requirements. No mandatory government safety testing. No enforced pauses. Voluntary compliance all the way down.

Meanwhile, Anthropic, OpenAI, and Google have each declined to commit to halting their own research. The brake pedal Clark is calling for? None of the major players are installing one voluntarily.

This isn’t hypocrisy so much as a structural problem. Individual companies can’t unilaterally slow down without ceding ground to competitors. That’s precisely why Clark is pushing the conversation toward governments and regulation—not because it’s comfortable, but because it’s the only lever with enough reach.

What This Means for the Workforce

Clark didn’t shy away from the economic dimension either.

AI agents—autonomous bots handling routine tasks—are already displacing roles that once required teams of engineers. Major tech companies have leaned into this narrative to justify mass layoffs. The math is blunt: if AI can do the work of hundreds of engineers, the headcount math changes fast.

But Clark offered a counterpoint that’s worth sitting with.

“There are open questions about whether AI systems can be truly creative… there is not really evidence for that yet.”

At Anthropic, he noted, the bottleneck isn’t engineering capacity—it’s good ideas. The systems can execute. They struggle to originate.

That’s a meaningful distinction for anyone mapping their career against an AI-saturated economy.

The Surprisingly Human Advice

Asked what a young person should do in a world that feels increasingly built for machines, Clark’s answer was almost defiantly analog.

Develop a hobby. Study liberal arts. Read broadly. Stay curious.

“People that are creative and can think broadly, people that read a lot, people that have interests are the ones most benefited by this,” he said. “Indulge in curiosity and it pays back in how you can use this technology.”

It’s not the answer you’d expect from someone building frontier AI. But it might be the most honest one available right now.

The Takeaway

Clark’s message isn’t “stop the machines.” It’s “build the infrastructure to steer them before steering becomes impossible.”

The gas pedal exists. It’s being floored. The brake pedal is a policy conversation that governments are moving toward slowly, companies are nudging cautiously, and the technology is outpacing entirely.

For founders, builders, and anyone choosing AI tools today—the ecosystem is accelerating whether the governance catches up or not. The smartest move isn’t to wait for the framework. It’s to understand what you’re working with, stay curious about where it’s going, and keep asking the questions that the systems themselves can’t yet think to ask.