Didn't work (yet): bots that failed to replace an app
Real, documented attempts where an AI agent didn't manage to replace an app or chore. Each one comes from a public test or write-up we read ourselves. Nothing is invented, and they're not rated.
Why list failures? Knowing where agents get stuck (login checks, payment security, fiddly forms) saves you an afternoon. Most of these are browser agents doing shopping or booking, which fits our ratings: browser-agent recipes are among our lowest-rated, while the steadiest swaps are scheduled automations with a human check.
Didn't work (yet)
Online grocery shop at Asda and Tesco with ChatGPT agent
Tried: ChatGPT agent · Would have replaced: Doing the online grocery shop yourself
What happened
A trade publication tried ChatGPT's agent mode on the UK's four biggest supermarkets. At Asda and Tesco it couldn't even build a basket: both sites' login checks stopped it, across multiple attempts and accounts. Only Sainsbury's and Morrisons let it complete the shop.
Why it failed
Retailer bot and login checks (Asda's Cloudflare check failed at log-in; Tesco showed an error page). OpenAI told the publication that some retailers block the agent automatically.
Retail Week ↗ · News test · September 2025 · read by us Oct 17, 2026
Didn't work (yet)
Fill a 16-item grocery cart with Operator
Tried: OpenAI Operator · Would have replaced: Ordering groceries online by hand
What happened
A writer gave Operator a real 16-item shopping list. It misread the list (adding unchecked items, reading “(5)” as “15”), then added only 13 of 16 items. On a second try it got logged out, became confused, and took more than 30 minutes with several interventions.
Why it failed
Misreading the list, logouts and confusion. The writer did the same order without AI in under 4 minutes.
Understanding AI ↗ · Newsletter test · June 2025 · read by us Oct 17, 2026
Didn't work (yet)
Shortlist 5 lamps into an Etsy cart with ChatGPT agent
Tried: ChatGPT agent · Would have replaced: Shopping around on Etsy yourself
What happened
A reviewer asked the agent to find 5 lamps under $200 and add them to the Etsy cart. It worked for 50 minutes and said it had added all five, about $825 in total. The reviewer's real Etsy cart was empty. When asked, it said it can't place orders or enter payment details.
Why it failed
The agent runs in its own virtual browser without the user's logins, so any cart it filled wasn't the user's. It also reported doing something it hadn't done for the user.
The Verge ↗ · News test · July 2025 · read by us Oct 17, 2026
Didn't work (yet)
Book flights directly with airlines using browser agents
Tried: Several browser agents and AI booking tools · Would have replaced: Booking flights on airline websites yourself
What happened
A consulting firm tried to book with major European and North American airlines and two big travel agencies, counting it a success if a tool reached the payment page. No tool did so reliably. Travel-agency sites did better than airline sites.
Why it failed
Browser agents stalled on date pickers and fare selectors, looping for 10–20 minutes before giving up. CAPTCHAs, two-factor checks, ID challenges, extras selection and session time-outs broke the runs, and people often had to step in.
Tried: ChatGPT agent · Would have replaced: Booking flights yourself
What happened
With a detailed prompt, the agent found the flight the author would actually book in 3 of 3 tests (each run took 10–30 minutes). But it couldn't finish the booking, so the author clicked through and booked it themselves.
Why it failed
EasyJet and British Airways blocked the agent, and the card-payment security check (3D Secure) stopped it every time. Search worked; booking didn't.