Why radiology is the shape of the second base camp.

We described the long-term thesis in our founding essay: medical superintelligence will arrive, regulation will follow necessity, and the decisive bottleneck is the deployment layer - what we call the harness. Clinical integrations, liability architecture, data standards, regulatory scaffolding, trust infrastructure. All of it has to exist before a model that can practice medicine reaches a patient who needs it.

A company cannot be built on a 5-15 year bet alone. You need revenue, you need customers, and you need to learn the landscape by operating inside it. So we climb in stages. Each stage is a base camp: a product that solves a real problem for real customers, earns its own revenue, and leaves a piece of the harness in the ground behind it.

The metaphor is deliberate. Base camps serve the summit expedition, but each one has to justify its own existence. A base camp that does not acclimatize the climbers, stock supplies, and hold its own against the weather is a liability, not an asset. The same is true for our products. Each one must stand as a business on its own terms. The strategic compounding into the harness is the bonus, not the excuse.

SignatureAPI is document infrastructure for healthcare. It handles the signing, verification, and compliance layer for clinical documents - the kind of work that every health system does manually, badly, and at scale. It is not glamorous. It does not involve frontier AI. It is the right first base camp for three reasons.

It teaches us healthcare operations without clinical risk. Document infrastructure sits in the administrative layer of a health system, not the clinical layer. A bug in a signature workflow is a compliance headache. A bug in a clinical AI system is a patient safety incident. Starting with documents means we learn how hospitals buy software, how their IT departments operate, how their compliance teams think, and how their procurement cycles work - all without the regulatory and liability complexity of touching clinical decisions. The tuition is lower.

It builds integration muscle.Getting SignatureAPI deployed inside a health system requires connecting to their document management systems, their credentialing workflows, their identity providers, and their audit infrastructure. Different from an EHR clinical integration, but the organizational muscles are the same: vendor security reviews, BAA negotiations, IT project management, go-live support. Every deployment teaches us something about the institution's architecture that will matter when we come back with a clinical product.

It generates revenue on its own terms. Health systems process millions of clinical documents per year. The market for document infrastructure in healthcare is real, measurable, and does not depend on the medical superintelligence thesis being correct. If we are wrong about the long-term bet, SignatureAPI is still a business. If we are right, it is the first piece of the harness.

We evaluated five candidate surfaces for the second base camp: opportunistic radiology screening, clinical documentation summarization, prior-authorization automation, prescription monitoring, and clinical trial matching. Radiology won on three criteria we weight most heavily.

Harness compounding. The harness for opportunistic findings - PACS integration, SaMD regulatory submissions, liability contracts for AI-assisted diagnosis, structured clinical output standards - transfers directly to every other clinical AI deployment surface. A PACS integration built for opportunistic screening works for any radiology AI application. An SaMD submission process built for a narrow classification model is the same process for any narrow classification model. A liability framework for AI-assisted diagnosis is the template for AI-assisted anything. The compounding rate into the harness is higher than any other candidate.

Deployment geometry. We wrote about this in detail in the radiology essay: the data already exists, the task is narrow and well-defined, the baseline is zero (nobody was screening for these findings), and the clinician stays in the loop by default. This is the easiest shape to deploy, which means it is the fastest path to production evidence - outcomes data, clinician feedback, institutional trust - that every subsequent deployment will need.

Market timing. The FDA has cleared over 700 radiology AI products. The clinical validation is done. The deployment infrastructure is missing. Hospitals have approved budgets for radiology AI but cannot find vendors who will own the full deployment stack: integration, liability, regulatory, monitoring, support. The market is waiting for the harness. We intend to provide it.

Intellectual honesty requires naming the conditions under which we would change course. Three scenarios would cause us to pick a different second base camp.

The reimbursement gap proves fatal. If health systems under value-based care and self-insured models will not pay for opportunistic screening without a CPT code, and CMS shows no movement toward creating one, the addressable market may be too small to build on. We think this is unlikely - the cost-avoidance math is compelling, and early conversations with risk-bearing systems have been encouraging - but it is the highest-probability failure mode.

A competitor ships the harness for radiology first. If one of the 700+ FDA-cleared radiology AI companies decides to build the full deployment stack - integration, liability, regulatory scaffolding, not just the model - and does it well, the opportunity cost of entering after them may be too high. We watch this closely. As of April 2026, none have. They are model companies, not deployment companies, and the gap between those two things is why we exist.

A faster-compounding surface emerges. If a regulatory shift (say, CMS mandating AI-assisted prior authorization review) creates a deployment surface with even higher harness transfer than radiology, we would consider resequencing. Strategy is a living document. The base-camp order is a current best judgment, not a commitment etched in stone.

The compounding is the whole point. Each base camp is chosen not just because it is a viable business but because the infrastructure it forces us to build carries forward.

SignatureAPI forces us to build: health system sales motion, vendor security review process, BAA and compliance infrastructure, institutional relationship capital, and a deployment operations team that knows how hospitals work. None of this is clinical. All of it is required for every clinical product we will ever ship.

Opportunistic radiology screening forces us to build: PACS and DICOM integration layer, SaMD regulatory submission capability, clinical liability contracts, structured clinical output standards, and pilot-site evidence that proves AI deployment works in production. These are the first five pieces of the harness that are specifically clinical. Every future deployment surface - prescribing, documentation, triage - will need some version of each.

The third base camp, which we have not yet named, will build on both. By the time we reach it, we will have institutional relationships, deployment operations, clinical integration capability, regulatory submission experience, and production outcomes data. The deployment gap that stops everyone else will be infrastructure we have already crossed.

This is the bet. Each base camp earns its own revenue and solves its own problem. Together, they build the harness that makes medical superintelligence deployable. The sequence matters because the compounding is directional: administrative infrastructure before clinical infrastructure, narrow clinical before broad clinical, imaging before decision-making. Each step makes the next step possible and cheaper.

We laid out the long view in the mountain top. This is how we climb.