We all want better health.

People are working on immortality treatments and brain uploads. I’m excited that they are – working on raising the ceiling of what’s possible in health – it makes me excited about the future we’ll get to live in.

At the same time, virtually no one, even in the world’s richest countries, has access to the ceiling of what’s possible today. For almost everyone, there’s not only a small gap – it’s vast.

The future of healthcare is already here. It’s just unevenly distributed.

Coagen – a new company I’ve been working on this past year – is motivated by two questions that I think are tractable to solve in the next 5-10 years:

  1. What if we could take the level of clinical workup you’d get at the best clinics in the world with an unlimited budget – and give it to everyone, anywhere, for free?
  2. What if we could make clinical trials 10x cheaper to run? How many more clinical trials would humanity run? 10x? 100x? 1000x? Jevons’ paradox for clinical trials. And what would this do to the effectiveness and price of the drugs that reach the market?

We build specialised AI models for clinical assessment, software for using them, and software they can use. Right now, this means an eCOA system for running and recording assessments, with AI that can review how a human conducted an assessment, produce its own clinical rating, or conduct the assessment itself. We’re starting with CNS disorders – everything to do with the brain.

A lot of this is really “boring”. I think today’s AI can already do much more clinical work than it’s being used for. The work is in proving it, integrating it, and getting the cost close to zero.

Coagen wants to be the most “boring” AI company for healthcare and clinical trials.

I think both problems — massively expanding the supply of expert clinical assessment and massively expanding the supply of clinical trials — should be tackled together.

In the future, the lines between clinical trials and clinical care will blur. People already join trials in part for the detailed assessments and close attention they can get, and to get better treatment options. Using healthcare as trial infrastructure can help fund healthcare, and tighter integration can help recruitment for trials. More of the trial process can eventually move “post-market”. In the future, most people with a condition will be diagnosed, and joining relevant research should be a routine part of healthcare.

I think both problems come, in large part, from an extreme shortage of expert human labour.

If we had 1000x more psychiatrists and 1000x more therapists (with their salaries covered), mental healthcare would look very different.

You can codify parts of best practice and turn them into software. With AI, many more parts can be turned into software. Then you can scale up best practice massively and cut the costs massively. This is an idea we take extremely seriously.

This applies directly to the supply of clinical assessment. And the nature of software means it’s global.

It also applies both indirectly and directly to the supply of clinical trials. A massively expanded supply of clinical workups in the broad population will make it easier, faster and cheaper to find the right patients for a trial. A large cost of trials are clinical workups at sites. And trials are limited by availability of trained trial staff: this will help in general, and it will help in particular in expanding trials to diverse geographic regions.

We want to replace Eroom’s law with Jevons’ paradox.

Cheaper trials mean more ambitious research programmes can be tested, instead of incrementalism. It means drugs for smaller populations can be tested, drugs for less common diseases, and for stratified patient populations. It means many more companies developing many more drug candidates in parallel – more competition between treatments, which can bring prices down.

Faster recruitment translates into faster time to market.

The pharma companies can make more money while drug prices fall. Clinical trials account for a large share of drug development costs. Making trials much cheaper can make drugs more profitable at lower prices. Lower prices can also expand the number of people who can afford a drug. Suppose a treatment reaches 100 million people instead of 100,000, at a hundredth the price, that’s 10x the revenue; and more than 10x the profit with a reduced cost base.

And 100 million people got the treatment.

Massively expanding the supply of clinical trials is even more urgent now that AI is being used to accelerate preclinical drug development. Trials are still the biggest bottleneck. If AI massively expands the supply of drug candidates, it will massively expand the demand for clinical trials, making trials even more of a bottleneck than they are today.

For those in the back: we need to massively expand the supply of clinical trials.

This is what Coagen is working on.

We largely do this by focusing on massively expanding the supply of expert clinical assessment. This includes both doing research and building products.

Our research focuses on:

  • Better datasets and benchmarks to measure human experts and AI systems.
  • Realtime multimodal AI models, and their integration with long-running agents.
  • Foundation models of behaviour.
  • Making these capabilities inexpensive to run, including local models.

Our products focus on:

  • Building the best interfaces to do work with our models.
  • Building the best interfaces for our models to do work.
  • Making “table stakes” software – which we think includes almost all software used in clinical trials today – free to use.

We are starting with Central Nervous System (CNS) disorders – everything to do with the brain. There is a large burden of disease, treatment options are often poor, and for many of these conditions a large part of the assessment is behavioural.