“Do you really know how to clean your hands? Without leaving the bits between your fingers?”
It was the last morning of the track cycling at the London 2012 Olympics, and Dave Brailsford, British Cycling’s performance director, was on BBC Breakfast. He had been asked about the “marginal gains” behind his team. “The whole principle came from the idea that if you broke down everything you could think of that goes into riding a bike, and then improved it by 1%, you will get a significant increase when you put them all together,” he said.
Some of it was fitness and training. Some of it sat at the edges: sleeping in the right position, taking the same pillow when away. Wash your hands properly, he said, and “you will get ill a little bit less.” The pillow idea, the BBC reported, apparently came from the Royal Ballet, and the hand-washing from talking to surgeons.
Britain won seven of the ten track golds, as it had in Beijing four years before. Beaten rivals wanted a better explanation. The French team’s boss suggested the British wheels couldn’t be the same ones her riders used. Brailsford told the French paper L’Equipe the wheels were “round.”
The case follows Brailsford’s road team, Team Sky, to a less comfortable place. In 2018 a committee of MPs said the team had used a medicine allowed under anti-doping rules to improve performance, crossing the ethical line Brailsford said he had drawn himself. The team and its rider denied it.
A habit, not a project
At Toyota, kaizen means continuous improvement, done by everyone. It sits beside “customer first” and “total participation” as one of the principles of the company’s quality management.
Marginal gains is the coach’s version: a performance director hunting for advantages everywhere, borrowing from ballet dancers and surgeons. Toyota’s version leans on the people who do the work. In 1950, according to Toyota’s in-house magazine, a labor dispute ended with a quarter of its workers leaving, and the company was short of money, equipment and goods. Around then Eiji Toyoda, visiting a Ford plant, saw a suggestion system that rewarded creative solutions on the factory floor. Toyota launched its own in 1951. It got 789 ideas in its first year.
In 2023 it got about 810,000, or 14.4 per person among the plant technical staff, the magazine reported. Most are small. Mitsuru Kawai, who started at Toyota’s training school after junior high and spent 61 years on the factory floor, gave an example: moving a trash can so you don’t have to walk around it. “Supervisors need to take such suggestions seriously, too,” he said. “Otherwise, that person will not make another.” The point, he said, is “to make our jobs easier.”
The procedure matters. In 1973 Toyota said that because it had been “carefully devised,” the share of suggestions adopted had risen from 29% in 1956 to 76% in 1972. Today you note a problem in your daily work, look for its cause, try a fix and check the result, and only then fill in the form. A suggestion arrives already tested.
When the count becomes the goal
Toyota’s own numbers carry a warning. Suggestions soared in the 1980s. Part of the reason, a Toyota kaizen specialist told the magazine, was competition between plant divisions, and because it “emphasized quantity over quality,” Toyota changed the system. Since 2019, rewards have been scored partly on the effect an improvement had.
A drive for gains also needs a boundary around it. The MPs’ report on Team Sky turned on exactly that: what the rules allowed, and where, in the committee’s view, an ethical line still ran.
The repair shop
Imagine a small bike repair shop where customers keep phoning to ask if their bike is ready. The mechanic writes it on the board. The owner tries one change for a week: a text message when the repair is done. The calls drop, so it stays. The next week, someone tired of opening the wrong spare-parts drawer relabels them all. None of this would make a business plan. After a year, the shop might work noticeably better, and no one could point to the day it changed.
Kaizen for an agent’s instructions
Claude Code, Anthropic’s coding agent, reads a file called CLAUDE.md at the start of every conversation: standing instructions for the project. Anthropic’s guide describes what amounts to a kaizen loop for it. Generate a starter file, then “refine over time.”“Treat CLAUDE.md like code: review it when things go wrong, prune it regularly, and test changes by observing whether Claude’s behavior actually shifts.”
Two parts of that match Toyota closely. A change is tested before it stays, like a suggestion tried before it is filed. And more is not better. The guide asks of each line, “Would removing this cause Claude to make mistakes?” If not, cut it, because “bloated CLAUDE.md files cause Claude to ignore your actual instructions.” Adding a rule after every mistake can turn into the 1980s suggestion count: more entries, not better work. Sometimes the improvement is a deletion.
Kawai’s trash can saved one walk around it. Taken seriously, it also told the person who moved it that the next idea would be worth making.