Dev.to AI πŸ€– Ai πŸ‘ 0 πŸ“– 1 min read

Korea's AI agents still have to make phone calls - a look at Pickle

Most agent products assume there is an API at the end of the task. In Korea, a lot of everyday errands still end with a person on the phone: changing a mobile plan, moving a clinic appointment, a refund that has no form.

Most agent products assume there is an API at the end of the task. In Korea, a lot of everyday errands still end with a person on the phone: changing a mobile plan, moving a clinic appointment, a refund that has no form. A browser agent can fill any form, but it cannot survive "please contact our customer center."

Pickle (피클) is a Korean app that takes the other route. It works websites in its own cloud browser, and when there is no web path it places the call itself and reports back what happened. Made by Pickle, Inc. (Y Combinator W25), iOS 17+ and Android, 18 and over.

The part worth watching, if you build agents:

  • The phone is a fallback, not the interface. You type or talk, and the agent decides whether this job is a form, a page to read, or a call. That decision is the product.
  • Capacity is the product too. In Korea it is free, with a daily usage limit that resets at 04:00 KST. Reset tickets top the limit back up, and inviting a friend gives both sides one.
  • Honest limits: the app and its help pages are Korean-first, and results still need checking. A hands-on English write-up from October found one of three researched venues wrong (address and price), so treat output as a draft.

Access: sign-up is invite-gated, and without an invite you land in a waitlist. My invite link, 30 slots, first come: https://pickle.com/i/RDKXK77G9ZRT

If you finish signing up through it, we become friends in the app and both of us get a usage reset ticket. The invite screen and the app are in Korean. I get a ticket too, which is the only reason I am posting it here.

Not official Pickle documentation (pickle.com is). If the link is closed, the slots are gone rather than the app.

πŸ“° Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes β€” full credit and traffic to the original publisher.