Published

AI in practice · Article

Robots in restaurants: what paid off and what went back to people

Adriano Meirinho 2 min read

In 2023, when we recorded episode 7 of /root, Alexandre Uehara had just come back from Japan and Korea with stories of robots that carry dishes to the table and others that clean it afterwards. I took the chance to ask the question a lot of people were asking at the time:

“So, guys, robotics plus AI, and society’s despair, jobs are going to disappear. If we replace journalists, marketers with artificial intelligence and replace waiters, other roles with robots, where is this society heading?” (free translation of the original in Portuguese)

Alexandre answered without hesitation. In his view, robots and automation take over the operational side, and creativity is hard to replace. ChatGPT has a certain creativity, he acknowledged, but someone needs to know how to ask the right questions. As he sees it, there will always be that operator, the person at the prompt who guides the machine.

The cases that have scaled in services since then follow that design, along a far less cinematic path than the movie humanoid. In December 2022, Skylark completed the rollout of 3,000 floor robots across about 2,100 restaurants in Japan, with the stated goal of taking the back and forth of dishes off the staff and leaving people to focus on service. At FamilyMart, Telexistence robots took over restocking drinks in the cold room, in a program announced for 300 stores; when the AI fails, a remote operator finishes the task. Going the other way, Henn na Hotel switched off about half of its 243 robots in 2019, because they created more work than they solved.

All three cases tell the same story. The machine pays off when the task is narrow and repetitive and someone is ready to take over the exception. When the exception has no owner, its cost eats the gain.

On November 9 I visited AGIBOT with the Endeavor delegation and came back with that impression reinforced. Robotics is moving fast, and what separates a project that pays off from one that ends up as decoration is still the choice of task and the design of the exception. In financial services, where AI agents already run entire process stages, I would start by mapping the edge cases and who is accountable for each one.

Sources: • Skylark Holdings, press release, 12/21/2022: https://corp.skylark.co.jp/Portals/0/images/news/press_release/2022/20221221_Robot.pdf • Skylark Holdings, DX report 2024: https://corp.skylark.co.jp/Portals/0/images/ir/section/2024/04_dx_E.pdf • FamilyMart, press release, 08/10/2022: https://www.family.co.jp/company/news_releases/2022/20220810_01.html • Telexistence, 08/10/2022: https://tx-inc.com/en/blog/2022/08/10/11737/ • Microsoft Customer Stories, Telexistence, 03/11/2025: https://www.microsoft.com/en/customers/story/22502-telexistence-inc-azure-synapse-analytics • The Verge, 01/15/2019: https://www.theverge.com/2019/1/15/18184198/japans-robot-hotel-lay-off-work-for-humans • NPR, 01/28/2019: https://www.npr.org/2019/01/28/689473868/this-time-humans-triumph-over-robots-as-they-take-back-hotel-jobs

Credits: /root, Celcoin’s podcast. Hosts: Adriano Meirinho and Thiago Zaninotti. Guest: Alexandre Uehara. Episode 7: IoT: Humano x Máquina, published in 2023. Excerpts from 9:08 to 11:20. Original video: https://youtu.be/v0MJuEyvLnw

Author

Adriano Meirinho · Co-founder & CAIO (Chief AI Officer) at Celcoin. About