This would be convenient! I wonder if you could have a fairly-decent first pass via a Chrome extension that hides all non-vegan items from the UI (or just greys them out).
You could probably use LLMs to do a decent first-pass on whether items are vegan. It’ll be obvious for many (e.g. vegetables, meat), and for non-obvious ones you could kick off a research agent who finds an up to date ingredient list or discussion thread. Then add the ability for users to correct classification mistakes and you’d probably be able to classify most foods quite accurately.
Then promote the Chrome extension via vegan magazines, influencers, veganuary, etc.
This is a thoughtful suggestion, and technically it’s quite feasible. The key issue, though, is effortlessness, which is the main behavioral bottleneck we’re trying to solve.
A Chrome extension (or app) is, by definition, an opt-in solution. It only helps people who are already motivated enough to discover it, install it, and keep using it. In practice, that means it mainly serves committed vegans—not the much larger group of new vegans, flexitarians, or people trying to reduce animal products, where most of the impact lies.
By contrast, a native vegan filter inside the supermarket UI is default-available, works on mobile, and requires zero setup. That difference matters a lot: friction at the point of purchase is one of the strongest predictors of vegan dropout.
We agree that LLMs, crowdsourcing, and user corrections are promising for data generation. In fact, those approaches are likely part of the backend solution. But as a delivery mechanism, browser extensions don’t scale to mainstream users in the way platform-level features do.
In short: the proposed extension could be a useful prototype or data-gathering tool, but it doesn’t solve the core problem—making vegan shopping effortless for everyone, by default.
I used a meal logging app once and the database it had was incredible, though not perfect. If the item had a barcode, the app had its nutritional data. So extension, agent, even an app with a camera can all work. Of course, I live in the US.
This would be convenient! I wonder if you could have a fairly-decent first pass via a Chrome extension that hides all non-vegan items from the UI (or just greys them out).
You could probably use LLMs to do a decent first-pass on whether items are vegan. It’ll be obvious for many (e.g. vegetables, meat), and for non-obvious ones you could kick off a research agent who finds an up to date ingredient list or discussion thread. Then add the ability for users to correct classification mistakes and you’d probably be able to classify most foods quite accurately.
Then promote the Chrome extension via vegan magazines, influencers, veganuary, etc.
This is a thoughtful suggestion, and technically it’s quite feasible. The key issue, though, is effortlessness, which is the main behavioral bottleneck we’re trying to solve.
A Chrome extension (or app) is, by definition, an opt-in solution. It only helps people who are already motivated enough to discover it, install it, and keep using it. In practice, that means it mainly serves committed vegans—not the much larger group of new vegans, flexitarians, or people trying to reduce animal products, where most of the impact lies.
By contrast, a native vegan filter inside the supermarket UI is default-available, works on mobile, and requires zero setup. That difference matters a lot: friction at the point of purchase is one of the strongest predictors of vegan dropout.
We agree that LLMs, crowdsourcing, and user corrections are promising for data generation. In fact, those approaches are likely part of the backend solution. But as a delivery mechanism, browser extensions don’t scale to mainstream users in the way platform-level features do.
In short: the proposed extension could be a useful prototype or data-gathering tool, but it doesn’t solve the core problem—making vegan shopping effortless for everyone, by default.
I used a meal logging app once and the database it had was incredible, though not perfect. If the item had a barcode, the app had its nutritional data. So extension, agent, even an app with a camera can all work. Of course, I live in the US.