Living as a Service: an operating system for human biology

From January 10 to January 30, 2026, we ran a study of a concept we call Living as a Service (LaaS). The research question: can an operating system for managing human biology be assembled from technology available today — and what has to be true for that to work. The method: a systematic review of 36 open sources — scientific publications, regulatory documents, and market data — followed by an architecture synthesis.

The fragmentation crisis

Healthcare has reached a bifurcation point. Genome sequencing is affordable, wearables record every heartbeat, continuous glucose monitors write metabolism in real time — and the value of all of it to a person remains negligible. The user drowns in disconnected reports: a PDF of genetic risks, a sleep chart from an app, a lab blood panel. None of these elements talks to the others.

Fragmentation produces direct conflicts: a genetic intolerance identified a year ago quietly contradicts a dietitian's fresh recommendation — and the person is the one who has to notice. Existing platforms solve data collection, not data synthesis.

Three generations of health management

The 2.0 → 3.0 shift is a paradigm change from the Quantified Self to the Automated Self: the system stops advising and starts executing.

The integration gap

Data about a person lives in three disconnected layers:

• The genomic layer — static: the system's boundary conditions, risks, and predispositions. A snapshot of probabilities, not an instruction for action.
• The phenotypic layer — periodic: labs and biomarkers that expire the moment the sample is taken.
• The behavioral layer — real time: sleep, activity, nutrition, environment — shallow without genomic context.

The LaaS innovation is a contextual engine computing derivatives between layers: how yesterday's behavior shifted the expression of risks given predispositions, and what exactly compensates that risk today. Formally — state synthesis:

S(t) = F( G, P(t), B(t), E(t) )

where G is the genome, P the phenotype, B behavior, E environment. Risk stops being a constant in a report and becomes a variable:

dPRS(t) = PRS · f( B(t), E(t) )

and health management becomes a trajectory-optimization problem:

u*(t) = argmin E[ Risk( S, u, horizon ) ]

The four-layer architecture

Layer 1 — foundation: genomics. The innovation isn't sequencing (now a commodity) but dynamic polygenic risk scores: a week of poor sleep and low activity recorded by sensors recalculates the instantaneous risk of a genetic predisposition realizing. Pharmacogenomics is a mandatory safety loop: every prescription is automatically checked against the genotype before it is written.

Layer 2 — the real-time sensorium. Hardware-agnostic aggregation of wearables through open APIs, continuous metabolic monitoring, and the home as a health hub: ambient data — air quality, temperature, light — explains what neither activity nor nutrition can.

Layer 3 — the computational core. The principled choice: a mechanistic digital twin instead of purely statistical models. Statistics predicts the future from "similar people"; a mechanistic model simulates the physiology of one specific person and lets the system play out what-if scenarios with mass Monte-Carlo simulation. The twin is adaptive: an injury or illness instantly rebuilds the strategy from performance optimization to recovery optimization.

Layer 4 — the execution engine. Large Action Models turn recommendations into execution: negotiating a doctor's slot, assembling a grocery basket against the nutrition plan, preparing the bedroom for sleep an hour before the calculated bedtime. The person keeps only the confirmation.

The operating model: a living strategy

The system's product is not a report but a living strategy: a dynamic document recompiled weekly, like a software release. A week of travel and poor sleep automatically turns the next week into a recovery week.

The physician is not replaced but amplified. They see an interpreted picture rather than raw labs; they receive signals only when a trajectory deviates (management by exception) — which lets one doctor supervise hundreds of patients; for narrow specialists the system automatically prepares a data package. The key metric shifts from "sick/healthy" to the pace of aging: epigenetic clocks make the speed of biological aging a measurable KPI, and the trajectory visual — "at the current lifestyle the chronic-disease threshold is crossed at age N; with corrections it moves years out" — becomes the strongest motivational instrument.

The psychology of engagement

Data is boring, and nobody reads PDFs. Published studies of digital future-self interventions show that visualizing one's future self sharply improves long-term health decision-making, overcoming the cognitive bias of temporal discounting. The second mechanism is hyper-personalization through the "biological loop": a recommendation grounded in a specific gene, marker, and context creates a sense of attention to detail no 15-minute appointment can match.

The market landscape

The market review showed mature solutions specializing: metabolic-disease management, ultra-premium diagnostics, platform-level data collection, hardware clinics. Each is strong in its own layer; the synthesis of layers and the execution loop remain open. That gap is what makes LaaS architecturally timely.

Regulation and ethics

The regulatory review produced three conclusions. First, the SaMD boundary: software that diagnoses or prescribes becomes a medical device with costly certification; the strategically correct start is the general-wellness category, with medical interventions passing only through a licensed physician who carries responsibility. Second, in Europe medical AI is classified as high-risk — the architecture must be explainable: the user and the physician must see why the system recommended what it did; black boxes are unacceptable in the medicine of the future. Third, privacy and data sovereignty: federated learning, where raw data never leaves the device, and the user's ownership of their own digital twin — including the right to take it and leave.

Study conclusions

1. Shift the focus from advice to automated execution — the agentic layer is technologically ready.
2. Build mechanistic modeling rather than plain statistics — the main technological barrier and the main defensible value.
3. Use future-self visualization as the primary motivation driver.
4. Build a hybrid model: AI as the architect, the living physician as verifier and guarantor of safety.

Health management is moving from apps to a human operating system. In a world where lifespan keeps growing, demand for managing its quality — healthspan — will only increase.

Economic models, the competitive analysis, applied implementations, and partner projects are covered in the full version of the study and remain outside this publication.