Shopping & prices: AI agent use cases

Price comparison sounds like a perfect agent job. In practice most retailers block bots, and search snippets go stale. These cases show what an agent can really see when it shops for you.

Every card shows two ratings: whether we reproduced the result (V2 reproduced, V1 partly, F failed, V0 not yet run) and how much account access the task needs, from L0 (public web only) to L4 (spends money or deletes). How we test.

1 case: 1 partly reproduced.