Sanctuary Health Manifesto

May 15, 2026

Healthcare should be abundant. There is no law of nature that says a person in the UK should wait eighteen months for surgery. These are not inevitable features of modern life. They are symptoms of systems that have become so burdened by their own complexity that they can no longer do the thing they were built to do: deliver care.

We started Sanctuary because we believe AI can change this and bring about an era of healthcare abundance.

To explain why we believe this, and why now, it helps to first understand what went wrong.

Diagnosis: Compounding Complexity

Healthcare is a system constrained by complexity that has compounded over time through well-intentioned intervention.

Roughly every five years, successive UK governments repeat a familiar cycle: consult the civil service to identify the system's main problems; receive a set of diagnoses, often accurate in isolation; design new plans, targets, incentives, and structures to address them; implement reforms over the course of a parliamentary term; then abandon or supersede them when the next government arrives. The process restarts.

The problem is, when viewed in isolation, each intervention appears reasonable. The people in Westminster are motivated by genuine failures, and they are designing their interventions with good intentions.

Healthcare is inherently complex because people are complex. The system manages human health – a biological and social phenomenon of extraordinary intricacy – under conditions where the worst-case outcome is death. This creates a profound asymmetry of risk: the cost of getting things wrong is morally, politically, and emotionally intolerable. This asymmetry causes a deep aversion to failure. In response, politicians and bureaucrats attempt to bound risk through rules, regulations, targets, and oversight. The system increasingly seeks to abstract away uncertainty by codifying behaviour however they can: specifying processes, defining acceptable actions, and reducing discretion wherever possible.

The NHS is not failing despite increasing regulation and oversight, but partly because of it. In complex systems, each attempt to reduce risk or impose control increases systemic complexity, which in turn generates new failures that demand further intervention. The result is a self-reinforcing cycle: management replaces judgement, bureaucracy replaces care, and stability is purchased at the cost of adaptability.

Processes only ever accumulate. The system is structurally biased toward invisible failure – slow deterioration, rationing, delay – over visible failure, where a specific identifiable error is caused by a specific, identifiable change. No one is fired for the ten-thousandth patient who waited eighteen months for surgery. But someone is fired when a new policy causes a single, traceable death.

Misallocated supply

The resulting problem from years of compounding regulation and complexity is that a huge % of the supply of healthcare is misallocated towards low-productivity work. While we were working in a GP practice in Manchester, we saw patient files transferred between clinics via a mailbag. It was one of the admin team's jobs to then scan the documents and create a new patient record at that practice.

Another doctor I spoke with said that when planning surgery, he would order the equipment he needed weeks in advance. Then, when arriving in the operating room, about 40% of the requested equipment would be present. The solution to this was to send a junior doctor running around the hospital, collating the necessary equipment.

Finally, the referral process between primary and secondary care, at least in the UK, can only be described as Chinese whispers with different doctors and administrative staff shuffling documents through digital equivalents of air tubes in the hope that something would be done on the other end.

All of this represents a misallocation of resources. Too many well-trained and high-value people are oriented toward doing low-value, burdensome work.

AI Will Increase the Supply of Healthcare

Our core belief is that AI can relieve healthcare of its administrative burden and, in doing so, allow the system to actually deliver care. This is what we call healthcare abundance.

Historically, software has forced conformity onto healthcare. Entrepreneurs and engineers would observe similar but subtly different clinical processes and build a single product to sell across the sector. The product would be good enough at solving a problem that it was worth buying, but it would force doctors, nurses, and patients to conform to what the software company had built.

The world's biggest EHR, Epic, brings genuine benefits: data centralisation, security, patient record access. But it also forces clinicians into doing a thousand things they didn't sign up for, navigating a platform full of features they don't want or need. The result is feature and process bloat that creates a massive burden on clinicians. The scale and complexity of these products create an asymmetry of power; once hospitals were using these systems of record, they couldn't leave them. The transfer cost in the short term was always way too high for hospitals to do anything about it. The result is a market that exhibits monopolistic tendencies; rent seeking software companies with very low customer satisfaction.

It was also very hard for new market entrants to catch up. A hospital running on a large EPR was never going to switch to a small start up doing one specific thing very well; the iterative lean startup approach simply doesn't work in this use case. The whole benefit of EPRs is the centralisation of systems into one piece of software. The only way to have displaced large EPRs was to build your own, which was an equally difficult task. There is so much learnt knowledge inside the EPR providers as well as 20 years of code. It would have taken 100s of millions of dollars just to get close.

The technical moat these companies had was very real. But we believe this is changing:

First, the cost of writing and deploying code is collapsing. Deploying and iterating tailored systems across many local contexts was once economically prohibitive but is now within reach. This does not eliminate complexity, but it makes it workable. We can ship more complex systems, faster, for 1/10 of the cost.

Second, AI agents can meaningfully increase healthcare supply. Agents represent the first technology that can truly imitate the ingenuity of human beings. It is not unimaginable that a series of AI agents could complete the core tasks that clinicians and staff are currently burdened with – booking, referral management, documentation, equipment coordination, coding, and reimbursement. They will likely soon be able to go beyond administration and into the clinical domain.

A second order effect of the above is that the SaaS business model is at risk. Guy's and St Thomas hospital procured Epic for 15 years for £500m. This could be a bad idea for a few reasons which are: it will soon be possible to rebuild an AI native version of Epic for 1/10th of that cost. Secondly, the switching cost of large systems of record is rapidly shrinking as the technical side can increasingly be managed by agents. We think this all leads to a more competitive market of software providers for healthcare.

For the first time since the invention of modern computing, it is possible to write, deploy, and scale customised software in healthcare – software that can run agents and models capable of completing a wide variety of historically human tasks, the way humans would. If we can aggressively deploy this model, we can move toward a world of abundant healthcare.

Sanctuary's Mission

We will unlock an era of abundant healthcare with AI.

We will achieve this by putting engineers where the problems are. Not in an office designing software for clinicians, but inside clinical settings, watching, learning, and building alongside the people who hold the system together every day. We will focus on the most difficult and highest-value problems we can find.

From there, we will build customised software solutions capable of handling the context dependencies and heterogeneous reality of modern healthcare. Over time, we will develop a core capability set that allows us to scale more efficiently without sacrificing the specificity that makes our solutions effective.

What does this look like

A new type of company. Historically, consulting companies have struggled to deliver transformative technologies because their incentives aren't set up to do so. Technology also falls outside of their core competencies, primarily focusing on operational improvements. SaaS companies, on the other hand, are not structurally designed for forward deployed engineering as hierarchical siloed organisations. We think a new company will emerge, one that has a core technology capability set that can be customised into healthcare settings; combining a problem focussed consultative approach with the technical ability and value of vertical SaaS.

We also believe this type of company is better for healthcare and state capacity more broadly. Our core value is in execution and implementation which is reflected in our business model. Most of our margin is captured in an implementation fee and one off license. The customer then owns their instance of the tool in perpetuity, where we can never take it away from them. Our ongoing revenue is made from value add services like maintenance, model updates, regulation management. Similar to Red Hat, we see moat in the core technology stack itself and so aligning the entire business's incentives to not derive compensation from that work.

The future of care will be delivered and defined by agents. These AIs require connected systems of data and autonomy to drive patients through care. Humans will want to be “cared” for by other humans because of our shared finitude. However, the medical decisions and processing will be handled by thousands of AIs. We see our work involving two main streams of work; firstly, connecting disparate systems of data to provide context. Building agents on top of those ontologies to do work.

At the moment we are currently doing this inside the existing infrastructure, however, I believe that the way to drive long term institutional change is to build a flywheel. These flywheels come from verticalising such that you deploy agents into your own systems of record and clinical settings. As I look forward, it is probable that Sanctuary one day owns and operates a hospital, in which we are deploying agents that do large longitudinal workflows. If this is the case I think it is likely that we can deliver care at a lower cost per patient than any NHS trust. We will pioneer a new model of healthcare, one that has AI at its centre and allows humans to focus on what they do best; caring for their fellow humans.