Two stories this week sit at opposite ends of the same question, and the question is not "is AI good" — it is who decides how AI shows up in your life.

At one end, Mark Zuckerberg. On Monday, Meta released a manifesto to go with its latest models, arguing that superintelligence should be available to everyone rather than concentrated inside a few companies, and that keeping powerful AI open is a safety measure, not a risk. Meta backed the words with releases: Muse Glimmer, a lightweight model under a permissive open licence that runs on a personal computer, and developer access to a more capable model, Muse Spark 1.2. The framing is abundance from the top down — put the models in everyone's hands and trust that distribution beats centralisation.

At the other end, a pharmacy in Vermont. Kinney Drugs spent the spring routing patients through an AI phone assistant it named "Burt." By this week, after hundreds of complaints about missed medications, wrong dosages, and calls that never named the drug, it was pulling Burt off the incoming lines and going back to a touch-tone menu. Nobody voted with a manifesto. They voted by refusing to use it.

Put the two next to each other and the week's actual signal comes into focus. The supply of AI — more models, more open, more capable, in more hands — is not the contested part anymore. The contested part is acceptance: whether the person on the other end of a specific deployment will tolerate it. And acceptance is being decided at the edge, by users, one interaction at a time — not by whoever ships the model.

The Meta side: openness as a strategy, and a bet

Zuckerberg's argument is not new, but the packaging is sharper. The pitch is that open models prevent a dangerous concentration of AI power, empower people directly, and are therefore the safer path. It is also, not incidentally, the strategy that suits the company that is behind on frontier proprietary models and ahead on distribution. Both things can be true. Open weights are a real public good and a competitive weapon aimed squarely at the closed labs.

What the manifesto cannot settle by itself is the part that actually determines whether "AI for everyone" means anything: what happens when one of those widely-distributed models is pointed at a real person with a real problem.

The Kinney side: where the models actually land

That is exactly what the Kinney Drugs episode is about. The company did not abandon AI on principle. Its fix is precise: inbound calls go back to touch-tone, but the assistant stays on for outbound refill texts — now opt-in rather than automatic. It kept AI where the stakes are low and the customer can say no, and pulled it from where the stakes are high and the customer was trapped.

That line — high-stakes-and-inescapable versus low-stakes-and-opt-in — is the practical version of the safety conversation the manifestos have in the abstract. A world with more capable models freely available is also a world where more Kinney Drugs decisions get made, by people with no lab, no red team, and no framework, discovering the boundary the way Kinney did: by crossing it in front of their customers.

What to watch

The open-vs-closed debate will keep generating manifestos, because it is where the strategy and the money are. But the more useful thing to track over the rest of 2026 is the acceptance line: the growing list of specific, named deployments that got walked back — not by regulators, and not by researchers, but by the users who were handed a permissively-licensed, technically-impressive system and simply declined to keep using it. Supply is solved. Acceptance is the whole game now, and it is being decided one refused phone call at a time.