Why Better Health Demands More than a Chatbot

Building on Labs AI and AI on the Hers platform, we are now rolling AI-native care out to Hims weight loss members, enabling them to partner with AI and their providers to achieve better health. 

The Closed Loop 

We’re transitioning our platform to AI-native care by building it around a closed loop: a system that remembers customers’ context, learns from outcomes, and continuously gets smarter and more valuable for customers and clinicians over time.

In previous telehealth models, a customer with nausea would need to report it manually, determine if it had a connection to their last dose increase, and communicate with support teams who couldn’t advise them. Then, they would get forwarded to a clinician, only to wait hours for a message back and have to explain their situation all over again. 

Our closed loop model is different, faster, and more seamless. When the customer chooses to use our AI tool and reaches out, the AI clinical engine knows their dose and treatment day, and that they haven’t reported this symptom before. It responds with an informed question about severity and either provides non-clinical direction or escalates straight to a provider. The customer now has an action plan and consistent follow-ups to check on their progress. The learnings from this outcome feeds back into the system, closing the loop and helping it get smarter for the next customer. 

This is the moat. While legacy telehealth struggles to iterate and foundation models are becoming commoditized, closed loop systems create advantages that benefit customers and are nearly impossible to recreate. 

From Advice to Action

General-purpose LLMs are built around advice. Our AI clinical engine is built around action.

That distinction carries real weight. If you're managing your health with a generic LLM, you first have to carry the burden of context. You’re re-explaining your history, your treatment, and your goals every time you open a new chat, all because you’re not sure what the model remembers and what it doesn’t. Then, you’re left with the burden of action, deciphering how to turn LLM advice into a next step, a scheduled appointment, a call to your doctor. When it comes to closing that loop – bringing all the pieces together – you’re on your own.

That’s the gap we’re closing. The AI clinical engine has persistent memory, a record that starts the moment a customer consents and opts in, and stays with them through intake, treatment planning, weigh-in results, and conversations with their Care Team. That shapes their app experience and includes information that is:

• Static, set at signup: name, state, stated goal, motivation, and prior attempts.
• Passive, updated automatically: medication, dose, treatment day, weigh-in trend, recent messages with their provider and Care Team.
• Growing, built session over session: context on why they’re doing this now, what success looks like to that person, the specific challenge they raised three conversations ago.
• Live: prior conversations in the current thread.

Every real outcome, whether a symptom resolved, a dose adjusted, or progress toward a goal, then feeds back into the overall system, sharpening it for the next customer while honoring each user’s privacy in the process. 

No general-purpose LLM has this. It can't. The ability to carry context and enable action only exists inside a platform that's actually delivering access to care. 

Accountable by Design

As the platform manages real customer interactions, an integrated safety layer helps keep AI responses safe, current, and personal. 

The AI clinical engine reviews each patient’s record against this layer, with our team monitoring outputs for anomalies and working to reduce potential errors or bias in the guidance it surfaces. Each AI response customers receive is:

• Gated: Before the system responds, it classifies each customer question to determine if it should be directed to AI at all. For a customer who needs clinical advice, the AI would route them to their Care Team instead of trying to answer their question on its own.‍
• Guardrailed: The AI operates within a defined scope, and is instructed not to respond to questions outside that scope. For example, if a customer asked for financial advice, the guardrails prevent the AI from providing it. ‍
• Grounded: We treat clinical guidelines as code, meaning clinician-written protocols are explicit rules the platform checks against when responding to customers. These are constantly monitored and updated in accordance with best practices by our clinicians. Responses must trace back to these guidelines. ‍
• Graded: An independent AI system grades each response for quality, and a sampling of responses is manually graded by humans as an additional layer of review. 

And, before any new version of our AI clinical engine goes live for customers, its outputs are run through a five-layer evaluation system, including offline red-teaming, a runtime protection classifier, and human-in-the-loop auditing at every step. Each layer is designed to catch a mistake before it affects someone's care.

Our engineers run simulated multi-turn customer conversations scored against pass/fail criteria our clinical team defines. These are full exchanges, not single prompts. And the bar isn't set by the engineers, it's set by the clinical leaders responsible for patient outcomes.

Every round of feedback gets converted into objective evals that drive the next iteration, whether from auditors reviewing live conversations, from drift signals in production, or from the Care Team. Every version of the system is accountable to the one before it.

AI Informs, Clinicians Decide

Clinicians remain an essential part of care. By providing context and direction on non-clinical questions, the AI clinical engine puts clinicians in the right place at the right time. It connects them to customers who need the help their specific expertise can provide, elevating time spent on clinical decisionmaking. 

That’s why the AI clinical engine’s routing logic is designed to hand off to a Care Team the moment a case needs human judgment. It ensures clinicians stay central in the care process. 

As we continue building our AI-native care platform, the clinical trust, closed-loop data, and a care operation we’ve built around each customer will evolve alongside it. That's the foundation, and it's what builds better, more intelligent care.