Esther Dyson on Onix: Trade-offs and Transparency

Esther Dyson wrote about Onix in Trade-offs and transparency, the July 28 edition of her Substack, In the pool with Esther. She attended our New York launch event last month, then spent time in conversation with one of the onixes. She came to it, in her own words, both as a reporter and as a prospective investor, and the piece reads like it. It is not a product review. It sets what we’re building against the AI-driven pill mills operating in the same market, and asks which business models can actually be trusted.

What she tested

Dyson talked with the onix built by Ashley Koff, RD, leading with electrolytes and then GLP-1s. Two things stood out to her, and only one of them is flattering.

The good one: the onix “kept listening and collecting facts for longer than any doctor has time for… and honestly, longer than I had time for.” That is the point. Time and patience are the scarcest things in a clinical encounter, and they’re the easiest thing for us to give away for free.

The other one: it kept steering her back to her own situation instead of engaging her broader questions about prevention versus treatment. She’s right that this is by design, since scope discipline is the whole architecture. She’s also right that it made the conversation less satisfying than the phone call she had with the real Ashley a day later. Worth sitting with rather than explaining away.

The constraints are the product

What I told her about how we bound the system:

Our concrete guardrails are no diagnosis, no prescribing, scope limited to what the human expert is expert in, only the expert’s own approved material as a source, and escalation to the person’s own provider for anything clinical.

Dyson’s summary of why that ordering matters is better than mine: “The trick is to start small, with a limited corpus, rather than to start large and try to control things with guardrails that are inevitably complex and focused on removing dangers that are hard to specify or anticipate.” You don’t get safety by bolting restraint onto something that already knows everything. You get it by never ingesting what the expert didn’t approve.

The incentives

She also asked about the money, which is the right question to ask anyone in healthcare:

Experts keep 70 percent. They set their own price. It is non-exclusive; they can leave any time and we delete their model and the data (the subscriber keeps their own text file of the conversations). If the experts do not make money, we do not make money. They are not our clients, they are our business partners.

We don’t sell drugs, supplements, or devices. Subscriptions are the only thing we sell, which means we have no reason to want anyone on a higher dose.

What’s still unresolved

The piece doesn’t let us off the hook, and shouldn’t. Ashley Koff is out of most people’s financial reach; an onix is meant to cost a fraction of a single appointment, but a subscription is still a subscription. When Dyson raised Tom’s Shoes, I told her we’ve been considering some kind of buy-one, give-one. That’s a direction, not an answer.

Her line for the broader risk is the one I keep coming back to: “The DNA of mission inevitably confronts the metabolism of money.” She closes by saying she’ll probably invest and then “fight like crazy” to keep us on mission. We’ll take both halves of that.

It’s the same test I’d apply to home robots and data brokers, pointed back at us. Don’t ask what a company has promised. Ask what it has made impossible.