When most people hear about AI in GLP-1 care, they picture a chatbot answering questions. That framing is not wrong, it is just incomplete and a little out of date.
Conversational AI does genuine work. An AI health assistant is useful as a round the clock support tool. A lot of patient questions do not need a clinician, a health coach, or even customer service. They need an immediate, accurate, supportive, and safe response. The usual human alternative is asynchronous, so a patient sends a message and waits a day or two for an answer. In the first few weeks of treatment, that delay alone can weaken the relationship between a patient and their provider.
Something I did not fully appreciate until we started evaluating patient-facing AI at scale is that many patients prefer AI for certain kinds of questions. Part of it is the immediacy. A bigger part is that they can ask anything without feeling judged. One patient told me she did not want to ask her dietitian about what she was eating, because she felt she should already know the answer. With AI, that barrier disappears.
The field has moved on in two ways that matter. The first is agents. Until recently, a conversation was more or less the only way to put generative AI in front of a patient. Agentic systems change that. They do not just talk, they take actions, pull data, trigger workflows, and hand off to other agents. Rather than one assistant trying to do everything, we build specialised agents for specific moments in the journey, whether that is goal setting, problem solving, or planning for what happens when things go wrong. Behaviour change is cumulative. It is small, consistent actions over time, and that maps well onto specialised agents working at specific moments.
The second shift is data. The genuinely new thing is not the conversation, it is the volume of data now flowing through these platforms and the ability to make sense of it close to real time. Weight, adherence, side effects, activity, sleep, engagement. Historically you could collect all of it and still not do much with it fast enough to help the person in front of you. Now you can, and you can learn from it, which interventions work for which patients, at which point, delivered with which cadence. That is where a lot of the real clinical and commercial value sits.
There is a useful distinction in the behaviour change literature between micro-engagement and macro-engagement. Micro-engagement is the in-app behaviour, the clicks, logins, and session time. Macro-engagement is the real-world behaviour the intervention is actually trying to change. The gap between the two is where most digital health tools struggle. Patients engage with the app but do not change what they do in their lives. Micro-engagement is a bridge to macro-engagement, not a substitute for it, and designing that bridge well is a lot of what we do.
So the chatbot is real, and it earns its place. The value has moved to the layer around it, the agents, the data, and the behavioural design that turns app activity into actual change.