.png)
There has always been talk about how independent AI should be in digital health. Mostly it was all about charts, scores, trend lines, risk summaries - outputs that require a human to look at them and decide what to do next.
AI agents are a different category. They do not just produce an output and wait. They perceive data, reason over it, and take action, triggering an alert, adjusting a recommendation, flagging a member for outreach, updating a care plan, without waiting to be asked. In a domain where continuous biometric data is already being generated around the clock by millions of wearable devices, that distinction is not a minor technical detail. It changes what is operationally possible.
A predictive model produces a number. It tells you that a member's cardiovascular risk score is elevated, or that a patient's HRV has declined significantly over the past two weeks. What happens next depends on whether a human looks at that output, interprets it correctly, and acts on it in time.
An AI agent closes that gap. It does not just identify a pattern, it actually responds to it. It can send a message, trigger a workflow, escalate to a clinician, or adjust a personalized recommendation, all based on the reasoning it applies to the data it is continuously observing.
The difference matters most in contexts where the volume of data is too high for human review at the individual level, where timing is critical, and where the cost of inaction is measurable. Digital health, built on continuous wearable data streams from large populations, sits squarely in all three.
The most immediate application of AI agents in digital health is continuous monitoring, watching biometric trends across a population and surfacing deterioration before it becomes a clinical event. At Thryve, this is something that we call the Prevention Pathway Guide.
A declining HRV trend over several weeks, a resting heart rate gradually increasing outside a member's personal baseline, sleep efficiency dropping consistently below a threshold, these are signals that a static dashboard will display, but that an agent can act on. Rather than waiting for a clinician or care manager to notice the pattern during a scheduled review, an agent can flag the member, trigger an outreach workflow, or escalate to a human reviewer in real time.
The clinical and financial value of this is significant. Early detection of cardiovascular deterioration, metabolic changes, or stress-related decline, before a hospitalization or emergency event, is where the largest gains in both outcomes and cost are available.
Engagement programs in digital health have a well-documented problem: generic messages sent on fixed schedules produce low engagement and minimal behavior change. An agent that observes when a member is most receptive, based on recent sleep quality, activity patterns, and historical response data, and delivers a targeted nudge at that specific moment operates on an entirely different level.
This applies across use cases:
The agent does not send a message because it is Tuesday. It sends a message because the data suggests now is the right moment. All of this can be done via our Member Connection Engine!
Static care plans are designed once and reviewed periodically. At Thryve, we provide our own care management model. For chronic disease management, rehabilitation, or wellness programs, this means a member may be following guidance that no longer reflects their current physiological state.
An AI agent with access to continuous wearable data can adjust recommendations in response to what the data shows today. A member recovering from cardiac rehabilitation whose recent activity and HRV data suggest they are progressing faster than expected can have their targets updated. A member whose sleep quality has deteriorated significantly can have their exercise intensity recommendations moderated until recovery improves. A patient whose resting heart rate has been elevated for several days can be prompted to rest rather than continue a planned training program.
This kind of responsiveness is what separates a personalized health program from a generic one, and it is operationally feasible at scale only when an agent is doing the monitoring and adjustment.
For clinical teams managing large patient panels, the bottleneck is rarely clinical knowledge, it is the time required to surface relevant information before a consultation or care decision. An AI agent that continuously monitors patient biometric data and prepares a concise summary of relevant trends, anomalies, and recent changes before a scheduled appointment changes how efficiently that clinical time is used.
Rather than asking a clinician to review weeks of raw data, the agent presents what matters: the trend in resting heart rate over the past month, the nights on which sleep was significantly disrupted, the days on which activity dropped sharply. The clinician makes the decision. The agent does the preparation.
At the population level, individual-level continuous monitoring is only useful if it can be prioritized. An insurer or health platform managing tens of thousands of members cannot have a human reviewer assess every member's biometric trend every week. But Thryve’s Health Risk Assessment model can.
Agents operating across a member population can continuously rank members by the significance of recent biometric changes, surface the individuals whose data warrants immediate attention, and deprioritize those whose trends are stable. This makes it possible to operate a genuinely proactive population health program, one where outreach is triggered by emerging risk rather than by scheduled review cycles or reactive claims data.
Remote clinical trials and observational research programs using wearables face a persistent operational challenge: ensuring that data collection is complete, that protocol deviations are detected promptly, and that adverse events are flagged in time to allow an appropriate response.
AI agents can automate much of this monitoring burden, tracking device wear time and flagging participants with incomplete data, detecting physiological readings that fall outside protocol-defined thresholds, and alerting trial coordinators to events that require review. This reduces the manual monitoring workload substantially while improving the reliability and completeness of the data collected.
An AI agent in digital health is only as reliable as the data it operates on and the infrastructure it runs within. Several requirements are non-negotiable.
An agent monitoring HRV across a population using mixed devices will encounter inconsistent metric definitions, varying sampling resolutions, and data gaps caused by device non-wear. Without a normalization layer that resolves these differences upstream, the agent's inputs are unreliable, and its outputs will reflect that.
When does an AI agent making health-related decisions become a medical device? The boundary between wellness and clinical decision support is not always clear, and the regulatory frameworks, MDR in Europe and FDA Software as a Medical Device guidance in the US, are still evolving. Organizations building agentic health applications need to assess regulatory exposure early in the design process.
The data that makes agents useful, continuous biometric monitoring at the individual level, is special category data under GDPR. Using it to trigger personalized health interventions requires explicit consent, clear purpose limitation, and infrastructure that enforces both.
The value an AI agent creates in digital health is directly proportional to the quality of the data it operates on. Inconsistent, unnormalized, or incomplete health data produces agents that make unreliable decisions, and in a health context, unreliable decisions have real consequences.
At Thryve, we build the infrastructure layer that agentic health applications depend on. Our API connects to 500+ wearables and health data sources and delivers normalized, structured biometric data through a single integration with GDPR-compliant handling, EU data residency, and consent management built in at the infrastructure level.
With the Thryve platform, you get:
If you are building AI-powered health applications and need a reliable data foundation, book a demo with Thryve to see how the infrastructure layer works in practice.
Paul Burggraf, co-founder and Chief Science Officer at Thryve, is the brain behind all health analytics at Thryve and drives our research partnerships with the German government and leading healthcare institutions. As an economical engineer turned strategy consultant, prior to Thryve, he built the foundational forecasting models for multi-billion investments of big utilities using complex system dynamics. Besides applying model analytics and analytical research to health sensors, he’s a guest lecturer at the Zurich University of Applied Sciences in the Life Science Master „Modelling of Complex Systems“
.png)