Mercari stopped work for a day: the CHRO-CAIO's account
その日、メルカリは「業務を止めた」――上司と部下が対立も、AI利用率100％をかなえたCAIOの奮闘記
B · Useful, limited ITmedia write-up of a conference talk by Mercari's CHRO and Chief AI Officer, with figures from the company's earnings materials; first-hand on organisation and pay, but company-reported numbers and no outcome measures.
Summary based on the full text·Original: about 6 minutes to read
In short
ITmedia Business Online (2 November 2026) reports a session from TOKIUM's online conference on 15 October by Toshiya Kimura, Mercari's executive officer, CHRO and Chief AI Officer, who led AI adoption as CTO. ChatGPT and Claude spread across Mercari in about half a year in 2025, yet business results did not follow, because only build speed rose. The company then set a policy of building agents in every business area and working on that basis. Uptake was uneven, and some managers stopped staff from spending time on agents, so Mercari halted normal work for a day. The talk also covers smaller teams, rewards for very high output, and telling staff honestly that jobs may disappear.
Key numbers
- 100%AI tool use across all staff at end of January 2026, per Mercari's FY June 2026 earnings materials
- 7.6xdevelopment output per engineer versus before AI (FY June 2024), per the same materials
- 1,000+non-engineering staff who stopped normal work for one "AI Agent Day" to build agents
- 8+ to 5-6team size in the "AI Pods" trial, premised on one person covering several areas with AI
- about 10xoutput of some individual engineers, described as about ten people's worth
As reported by the source.
Japanese·Web page·opens itmedia.co.jp
What it reports
- Kimura's view: speeding up building alone does not raise output; planning, development and every check from upstream to downstream must be turned into agents for the company to change.
- With about 2,000 staff, some did not know what to do and a minority of managers blocked agent work. He repeated at fortnightly all-hands that time spent building agents is not wasted, but "no time" complaints persisted.
- "AI Agent Day": 1,000+ non-engineers stopped work for a full day, learned agent building in about an hour in the morning, built agents for their own work in teams and presented them. Participants said their thinking changed that day.
- "AI Pods": teams cut from 8 or more to 5-6, on the hypothesis that people crossing areas with AI means fewer checks and faster decisions. Crossing within a job family (backend developers also building web and app) went well; crossing job families was harder, for example engineers could not fully judge an agent's compliance findings.
- Seniors who are good at delegating to people are also good at delegating to agents, he says.
- Rewards: base-pay expectations should rise in stages, with incentives for high performers for now; today's "10x" may be normal in a year or two, so redesign must start now. Mercari's August earnings materials state a plan to redesign AI implementation and evaluation together.
- Funding raises: management meetings frequently balance infrastructure cost, AI usage (token) cost and personnel cost, since pay cannot rise without revenue growth. He moved from CTO to CHRO to bring knowledge of how work is changing into HR design; the head of HR should co-design with whichever function leads change.
- An internal natural-language data agent, "Socrates", lets people query databases without SQL; analysts who handled requests now propose business direction.
How they got it
A journalist's summary of a conference session, with headline figures taken from Mercari's published earnings materials.
Limits
- Company-reported figures; "development output" is not defined in the article.
- No data on business outcomes, quality, attrition or pay changes.
- Edited talk write-up; no independent reporting.
Why it matters for companies
The account is unusually frank about the middle-management blockage: a policy and repeated messaging did not free time, a company-wide stoppage did. The HR points are the durable part: plan for productivity gaps of several times between people, decide whether to pay through base or incentive, and fund it by managing token, infrastructure and personnel costs together. Pairing HR with the function driving change, rather than leaving HR to react, is a design choice other firms can copy.
Key passages in translation
Japanese original
ものづくりのスピードだけが上がっても、成果はなかなか増えない。
Even if only the speed of building goes up, results hardly increase.
Japanese original
部下への仕事の頼み方がうまいシニア層は、エージェントへの任せ方もうまい。
Seniors who are good at handing work to their reports are also good at handing work to agents.
Japanese original
今の『10倍』は、来年や再来年には当たり前になっている可能性が高い。
Today's "10x" is likely to be the norm next year or the year after.
Summary based on the full text. Read the original for the full argument and data.Read the original on itmedia.co.jp