By the Wellness A\ team
Every Thursday evening, a cardiologist in Miami opens a queue of twelve cases. None of them are her patients. Each one is a conversation between a model and a clinician somewhere in our network, consented and stripped of anything that could identify a person, waiting for her verdict.
She is one of several hundred practising specialists on Frontier, the expert network behind the platform. The work looks nothing like the data-labelling farms that trained the last generation of AI. It looks like medicine: read the case, interrogate the reasoning, write down where it holds and where it breaks.
The unit of work on Frontier is disagreement. When the model’s suggested plan differs from what the reviewing specialist would do, that difference is the most valuable artefact we can collect. It is written up, sourced against current guidance, and folded into the next evaluation cycle.
“This is the closest thing to attending-level reasoning I have done outside the ward, and it actually changes what the model does next week.”
Every refresh is benchmarked against the previous one on a fixed battery of clinical evaluations before it ships. If a version does not clear the baseline on safety-critical categories, it does not ship, whatever its other numbers look like.
Medicine moves. Guidance changes quarterly, practice changes with it, and a reviewer who left the clinic five years ago is reviewing against a memory. We require active practice for the same reason hospitals require revalidation: the standard you hold others to has to be the standard you work to yourself.
Frontier pays for that judgment, remotely and flexibly, in hours that fit around clinic. If that sounds like your kind of work, we would like to meet you.