# #644 ベテランが20年かけたものを入社2年目の若手が1時間でつくった…企業のAI研修の現場で起きていること

- Show: プレジデント音声版
- Published: 2026-10-26
- Canonical: https://podc.club/en/episode/1525273315_464c381677dc.html
- Audio: https://traffic.omny.fm/d/clips/a10fab0c-08e4-4ce4-abed-aeb601152bcc/fa8d9e0c-1e7b-4e5e-81c9-aebc0016b5a2/d35d9dd9-f1a0-44d0-af94-b4e8001f6b1d/audio.mp3?utm_source=Podcast&in_playlist=2bdab59e-ba3e-4897-b598-aebc0016b5b0
- Source: Edited from a machine transcript of the first 30 minutes of the episode. Not an official summary.

A Stanford-MIT study of 5,179 workers found that AI tools raised productivity by 34% for novices but only a few percent for veterans, leading researchers to conclude that AI lifts beginners' starting points rather than making experts more expert. Generative-AI evangelist Sato Suguru (佐藤すぐる) explores that finding—and the hidden trap of what workers do with the time they save—in his book 'Claude Shigoto-jutsu,' discussed here by PRESIDENT magazine editor-in-chief Hoshino Takahiko (星野孝彦).

## What this episode answers

### Who benefits more from AI—veterans or novices?

Novices benefit far more: in a study of 5,179 people, overall productivity rose 14%, but less-experienced workers improved by 34% while veterans improved by only a few percent. A novice who used AI for two months matched the output of a non-AI worker with six months of experience.

### What happened when a veteran and a second-year employee tackled the same AI task at a manufacturing-industry training session?

The second-year employee used Claude to produce a ready-to-use departmental operations manual in just one hour, while the 20-year veteran found AI more trouble than it was worth because his own methods were already ingrained. The veteran acknowledged, 'He put into form in one hour what took me 20 years to build in my head.'

### What does Harvard Business School's 'Jagged Frontier' concept reveal?

The Jagged Frontier shows that AI improves output quality by 40% on tasks it handles well, but actually worsens accuracy on tasks it handles poorly. The author argues that the critical skill of the AI era is therefore not knowing how to use AI, but knowing what to hand to AI and what to do yourself.

### Why is it so hard to redirect AI-freed time toward deeper work?

Organizations naturally fill any surplus time with new routine tasks—what sociologist Max Weber called the 'bureaucratic trap,' where work generates more work. Field reports consistently show that time freed by AI gets consumed by a different set of routine tasks, leaving less time for deep work than before.

### What concerns has Bill Gates voiced about AI?

In a New York Times podcast interview, Gates expressed strong concern that AI robots could displace human jobs, going so far as to say they should be subject to the same social-insurance contributions and taxes as human workers. He also conceded, with some resignation, that data centers will be built regardless of opposition, and seemed troubled by the irreversibility of technological progress.

## Article

## AI Helps Beginners Most—Stanford × MIT Research

In 2023, a joint Stanford University and MIT research team conducted a large-scale study of 5,179 customer-support workers. AI tools raised overall productivity by 14%, but the breakdown was striking: less-experienced workers improved by 34%, while seasoned veterans gained only a few percent. The researchers also found that a novice who used AI for two months performed at the same level as an experienced worker who had spent six months on the job without AI. Their conclusion: AI is democratizing the know-how of top performers.

In other words, AI is not a tool that makes capable people more capable—it is a tool that raises the floor for people who are not yet capable. Those who feel they are weak with AI may, paradoxically, have the most room to grow.

## The '20 Years vs. One Hour' Shock in Manufacturing Training

Author Sato Suguru witnessed exactly this dynamic during a training session at a manufacturing company. A 20-year veteran and a second-year employee were given the same exercise: have Claude organize their own job duties into an operations manual. The veteran, whose methods were long established, felt that handing the task to AI created more friction than it removed. The second-year employee, whose habits were not yet fixed, accepted Claude's output openly and finished a department-ready manual in one hour. At the end of the session the veteran said quietly, 'He put into form in one hour what took me 20 years to build in my head. It stings, but that's reality.'

## Drawing the Line Between What AI Does and What Humans Do

Harvard Business School research introduced the concept of the 'Jagged Frontier'—the uneven boundary between what AI handles well and what it handles poorly. On tasks within its strengths, AI improved output quality by 40%; on tasks outside them, accuracy actually declined. The implication is that the defining skill of the AI era is not technical proficiency with AI tools, but judgment: knowing what to delegate to AI and what to keep for yourself.

First drafts of documents, data aggregation, and routine analysis fall on AI's side of the line. Decisions involving human relationships, ethical judgment, and creative responses to unprecedented problems remain on the human side. At a training session for an educational institution, a teacher asked whether it was acceptable to have Claude write all student feedback comments. The author's answer was 'half yes, half no.' Polishing sentence structure and correcting typos can go to Claude; deciding how to assess a particular student's effort and what goal to recommend next is something only the teacher can do. A few weeks later the teacher reported back: 'The time I spend on feedback is down to a third. But the content is actually richer than before.'

## What You Do With the Freed Time Is the Biggest Challenge

Sato proposes a '30-minute daily AI investment' in his training sessions. For the first two weeks, use Claude for email and document writing; saving five minutes per message across ten messages a day yields a 50-minute daily gain. The next two weeks are spent setting up a department-specific environment using Claude's Projects feature. After one month, 30 minutes to an hour per day should open up—time ideally redirected toward strategic planning, team management, and deeper customer relationships.

Editor-in-chief Hoshino identifies a major pitfall here. As sociologist Max Weber observed with the 'bureaucratic trap,' organizations have a natural tendency to generate new work to fill any available capacity. Fieldwork consistently shows that even when AI streamlines routine tasks, the freed time gets absorbed by a different set of routine tasks, and time for deep work actually shrinks. Sato stresses that to avoid the 'efficiency up, results unchanged' trap, workers must consciously decide in advance how they will use the time AI gives back to them.

## Bill Gates's Wariness and the Attitude of Finding AI Exciting

In a New York Times podcast interview, Bill Gates voiced serious concern about AI robots displacing human jobs, going as far as to say they should be taxed and assessed the same social-security contributions as human workers. At the same time, he showed a degree of resignation, noting that data centers will be built somewhere regardless of who objects, and expressed unease at the irreversibility of technological progress. Hoshino described Gates as 'torn between the benefits and the risks of the technology.'

Sato himself closes on an upbeat note: 'I look at material from six months ago and have to smile because it's already obsolete—and I find that enjoyable.' Just as with the spread of electricity, the automobile, and the internet, living in real time through an era in which a tool like AI is evolving at this speed is, he says, a rare experience in human history.

## Takeaways

- AI raises the floor for beginners far more than it raises the ceiling for experts—novices saw a 34% productivity gain versus a few percent for veterans—so those who feel least confident with AI have the most to gain.
- The core skill of the AI era is not knowing how to use AI but knowing what to hand to it and what to keep for yourself, with final accountability always remaining with the human.
- Time freed by AI must be consciously directed toward deeper work; without that intention, Weber's bureaucratic trap ensures new routine tasks will fill the gap and deep-work time will shrink, not grow.