The invention layer

A cardiologist in Poland built a patient-facing AI tool in seven days. Between shifts. He placed third at Anthropic's hackathon, out of 13,000 applicants. He's not a developer. He's a doctor who understood the problem so well that the code was the easy part.

This is worth pausing on. Not because it's a heartwarming story about a clever clinician, but because it signals something structural: the bottleneck in healthcare AI has flipped.

For three years, the sector's default question has been "how do we use AI to do what we already do, faster?" Scribes that document notes. Agents that file prior authorisations. Schedulers that route patients. All useful. All firmly inside the automation layer: taking existing workflows and shaving minutes off them.

The most important AI products in healthcare haven't been built yet, because the people closest to the problem have never had the tools to build them.

That's changing. Healthcare.digital calls it the birth of the "clinician developer": someone who can generate, debug, and deploy a functional application through a conversational loop with an AI coding assistant. No dev team. No six-month roadmap. No procurement cycle.

The implications run deeper than one hackathon winner. Insilico Medicine and Eli Lilly just published a framework in ACS Central Science for what they call "Prompt to Drug": fully autonomous, AI-orchestrated drug discovery from target identification to clinical planning. An open-source tool called celltype-cli lets researchers run multi-step drug discovery analyses in natural language. AI-discovered drugs are hitting Phase I success rates of 80 to 90 percent, nearly double traditional benchmarks.

Meanwhile, the US government is funding an autonomous AI cardiologist through ARPA-H. Utah is already refilling prescriptions via AI for $4. And a16z's Infinite Healthcare thesis argues that when care capacity becomes effectively unlimited, discrete visits stop being the unit of care altogether.

None of this is automation. It's invention. The difference matters.

🤖 Automation asks: how do we document the visit faster? 💡 Invention asks: how do we prevent the visit? Automation optimises the workflow. Invention questions whether the workflow should exist. The first is safe, measurable, and already happening. The second is where the value lives, and it needs domain experts, not just engineers.

One reply to the cardiologist's story put it plainly: "Domain knowledge is now the scarce resource, not engineering capacity."

If the people who understand healthcare best can now build the tools they've always wished existed, the question isn't whether the invention layer gets built. It's who builds it first.

The healthcare executives still asking "How do we use AI to speed up what we already do?" are solving the wrong problem. The ones who'll lead built something new last weekend.

💥 May this inspire you to build the tool you've been waiting for someone else to make.