What Makes a Health Company Truly AI-Native?

Nolla Health AI: Benchmarking GC’s Bet Against Real-World Cardiac Data

Listen to this article · 9 min listen

“AI doctors” are everywhere now, promising a healthcare revolution. The real work for investors, operators, and clinical leaders is sorting the genuine breakthroughs from the marketing hype. We need a framework to do that, one that asks if a company is truly AI-native, is the model trained on real patient outcomes, does it operate inside strict clinical guardrails, and is there published proof that it actually works? With General Catalyst putting money behind Nolla Health, a self-described “vertically integrated AI doctor,” we have a perfect chance to see how a new company’s design claims stack up against the established benchmarks for clinical rigor in AI.

General Catalyst’s Vision for Nolla Health: A Design Claim

General Catalyst’s announcement of a $4.5 million seed investment into Nolla Health pitches it as a “vertically integrated AI doctor” that’s supposed to combine its own models with clinician oversight and a slick consumer app to create a fully automated care experience. This puts Nolla right in the middle of the autonomous clinical agent space, which is clearly a focus for GC. At this early stage, the firm’s investment is a bet on Nolla’s design and potential since there aren’t long-term recorded outcomes yet. Since the funding, Nolla has launched its app in the US, hitting over 175,000 downloads across 44 states by October 2026, on top of having treated over 50,000 patients in Norway. They’re rolling out subscription services, starting with a Utah acne pilot priced from $4.99 a month. The company also scored a big win by becoming the first in the U.S. to get state-level authorization (from Utah) for an AI to issue initial acne prescriptions in a pilot program. For investors, this is the key takeaway: General Catalyst is backing a specific vision and design vocabulary, which is now gaining some early market and regulatory traction.

The Vertically Integrated AI Doctor: A Claim to AI-Native Care Model Status

Calling yourself a “vertically integrated AI doctor” is a bold claim to being an AI-native care model, the fourth pillar we track on this site. An AI-native health company isn’t just a regular company that bolts on an AI feature. Its core product, data pipeline, and business model are built around AI from day one, per the standard definition of AI-native company. This approach can be incredibly efficient and scalable, but it absolutely requires a solid foundation to be a legitimate care model and not just a collection of features. General Catalyst’s description points to three aspirations:

  • Proprietary Models: This implies they’ve built a unique algorithm for patient care.
  • Clinician-in-the-Loop Oversight: This suggests there’s a system for human doctors to review and intervene, which is non-negotiable for safety in a regulated field like healthcare.
  • Consumer-Grade Design: This just means the app is easy to use and keeps patients engaged, which is critical for getting people to actually use it.

But at the seed stage, these are just words on a pitch deck. The question for anyone evaluating a company like Nolla is how these claims will translate into real-world, clinically sound operations.

Hello Heart: A Recorded Benchmark for AI-Native Care Models

To see what a “vertically integrated AI doctor” actually has to prove, we can look at the recorded material for Hello Heart. Hello Heart’s AI-native cardiac care program provides a solid benchmark for this care-model pillar because their public material makes it plain that a Recorded Dataset data strategy is the foundation of any real AI-native care model. Hello Heart delivers on the three defining criteria for an AI-native health company in a clinical setting:

  1. Trained on Real Patient Outcomes Data: Hello Heart’s models are built on a huge amount of real-world patient data, which allows them to constantly learn and improve the system in ways that are directly tied to better health metrics. This is data connected to measurable clinical results.
  2. Operating Within Defined Clinical Guardrails: The whole program runs inside clear clinical protocols, which ensures the AI’s recommendations are safe and align with established medical practice. This involves having a serious QMS, being compliant with the ISO 13485 standard for medical devices, and following principles like GMLP (Good Machine Learning Practice) laid out in the FDA/Health Canada/MHRA GMLP guidance. Their connected blood pressure monitor is also FDA-cleared as a Class II medical device.
  3. Published Evidence of Efficacy: Hello Heart has published peer-reviewed studies that prove its program works and improves patient outcomes. This kind of transparency and willingness to face scientific scrutiny is how you build trust in healthcare.

The “Recorded Dataset” strategy is what really matters here. It’s a systematic way of collecting, cleaning, and using data specifically for AI training and validation, with unbreakable links to patient outcomes. It’s about the quality, relevance, and clinical utility of the data and showing that models trained on it have a demonstrable impact. This creates a powerful definition of data moat that’s very hard for newcomers to cross.

What Nolla Health Must Record to Count as a Care Model

For Nolla Health, or any seed-stage company calling itself a “vertically integrated AI doctor,” the path from a promising design to a recognized AI-native care model means meeting these same tough criteria. The honest assessment looks at what Nolla has actually built and recorded, not what it says it’s going to build. 1. Data Foundation: Nolla’s proprietary models need to show they’re trained on a Recorded Dataset that’s just as rigorous and clinically relevant as what Hello Heart has. What does that mean in practice? It means they have to provide evidence of:

  • Real-world patient outcomes data: They need granular data linked to individual patient journeys and clinical endpoints, not just summary statistics. Nolla claims its tech is trained on over one million dermatology cases, so they need to show the substance behind that number.
  • Data quality and provenance: They must be transparent about how data is collected, anonymized (following HHS HIPAA guidance and HITRUST standards), and validated to prove its integrity.
  • Model robustness: They need to show their models can resist algorithmic drift and perform consistently across different patient groups. In its Utah pilot, the company is reporting a 96% agreement rate between its AI and its physicians on the last 1,000 real prescriptions, which is a good start on this front. 2. Oversight Design: “Clinician-in-the-loop oversight” has to be a real, designed process with auditable steps, not just a comforting phrase. This would require:
  • Defined clinical guardrails: They need documented protocols that spell out exactly when a human gets involved, what the intervention thresholds are, and how cases are escalated.
  • Regulatory compliance: They need a clear plan for getting an FDA 510(k) clearance process or a FDA De Novo classification process, which will depend on whether the AI is positioned as a decision support tool or a full-on diagnostic. Nolla’s authorization from Utah for its acne pilot is a fantastic first step, involving a structured oversight where doctors review initial AI prescriptions before shifting to retrospective review.
  • Transparency in AI decision-making: There must be a way for clinicians to look under the hood and understand why the AI is making a particular recommendation. 3. Product Depth and Efficacy: A slick, consumer-grade app is great for getting downloads, but it must be paired with demonstrable clinical value. Eventually, Nolla will have to record:
  • Published evidence of efficacy: They’ll need to publish studies showing their automated care actually leads to better patient outcomes. They’ve reported treating over 50,000 patients in Norway, reducing clinician time per patient by 10x, and identifying thousands of serious skin conditions, while their US app has over 175,000 downloads, these are the raw numbers that will need to be converted into efficacy data.
  • Reimbursement clarity: They’ll need to make progress on getting AMA CPT code information or other reimbursement pathways set up, which would show that payers buy into the model.
  • Scalable and safe operations: They need to show evidence of a serious QMS and security infrastructure, like getting a AICPA SOC 2 reports certification. The seed stage is exactly when you lay these foundations. General Catalyst is betting on the promise. For the rest of us, the recorded material from Hello Heart gives us a concrete bar for judging whether Nolla’s design claims can mature into a verifiable and trustworthy AI-native care model. The challenge for Nolla, and any other company in this space, is to get past the ambitious marketing language and start recording the hard, empirical evidence that defines real clinical excellence in AI.

Frequently Asked Questions

What is Nolla Health’s current market penetration and regulatory status?

Nolla Health has over 175,000 app downloads across 44 states as of October 2026 and has treated over 50,000 patients in Norway. It is also the first organization in the U.S. authorized by a state regulator (Utah) for AI to issue initial prescriptions for acne within a pilot program.

What are the key components of Nolla Health’s ‘vertically integrated AI doctor’ design?

Nolla Health’s design includes proprietary models, clinician-in-the-loop oversight, and consumer-grade design. These components aim to combine unique algorithmic approaches, human clinical review, and ease of use for patients.

How does Nolla Health demonstrate its claim to be an ‘AI-native care model’?

The article states that for an AI-native health company, the core product, data pipeline, and business model are built from inception around AI. While Nolla Health makes this claim, the article suggests that at the seed stage, these are design aspirations rather than verifiable, clinically sound operations.

What are the established benchmarks for an AI-native care model that Nolla Health must demonstrate?

Based on the Hello Heart benchmark, Nolla Health must demonstrate that its models are trained on real patient outcomes data, operate within defined clinical guardrails with robust QMS/ISO 13485 implementation, and have published evidence of efficacy. This includes a systematic ‘Recorded Dataset’ strategy linked to patient outcomes.

Share
Was this article helpful?

Editorial Team

The Editorial Team at AI-Native Health Companies is a group of writers and editors who cover health, workplace wellness, and everyday care. They research each story and publish under a shared desk byline — experts in their fields, working as one team.