The One Who Saw a Banana and the One Who Tasted It

https://youtu.be/Gk-9Fd2mEnI

In a 1992 MIT Sloan lecture, Steve Jobs criticized consultants with his “picture of a banana” story. You can look broadly across many companies, he said, but unless you execute your own judgment for years and take responsibility for the results and the failures, your experience stays thin. He put it as: “You never get three-dimensional.” However many pictures of bananas you look at, it is not the same as tasting one.

I think this metaphor applies quite precisely to Korea's talent structure today.

Korean education's achievements are themselves remarkable. According to the OECD, as of 2025 Korea's tertiary attainment rate among 25-to-34-year-olds is 71%, among the highest in the OECD. Graduate education has economic value too: in Korea, holders of master's and doctoral degrees have an employment rate 7 percentage points higher and incomes 33% higher than bachelor's graduates. So this is not an argument that university or graduate school is unnecessary.

The problem is how accurately credentials represent ability.

The OECD has pointed out that Korean secondary education is excessively focused on university entrance exams, and that competencies that are relatively hard to measure — creativity, critical thinking, collaboration, communication — can be sacrificed to those easy to measure, like math, Korean, and English. It also analyzed that the fierce competition for top universities raises the burden on students and households, and can push students to choose majors by admission odds rather than aptitude.

The labor market shows similar traces. According to the OECD's 2023 Survey of Adult Skills, 31% of Korean workers held qualifications above their job's requirements, against an OECD average of 23%. The share whose field of study does not match their actual work was 49% in Korea versus an OECD average of 38%. Another OECD analysis found Korea's mismatch between educational attainment and job requirements at 44%, above the OECD-area average of 35%.

There are more direct criticisms as well. The OECD recently flagged the structure of the Korean labor market in which academic credentials and institutional prestige weigh more heavily than practical competence. Earlier OECD research on Korea's over-qualification phenomenon likewise pointed to the strong emphasis individuals and firms place on educational credentials, the preference for university graduates, and the gap between education and the competencies companies actually require.

So the problem is not that Koreans study too much.

It is that we evaluate with great precision how well someone can describe a picture of a banana, while the ability to actually eat one, sell one, let one rot, and pick better next time remains comparatively hard to evaluate.

Take business as an example.

Study SaaS and you can explain the definition of subscription software. But sell SaaS yourself and you learn why customers drop off right before payment, why retention collapses, why one small feature keeps generating operating costs.

SI, too, can be defined in a book as building custom systems for clients. Do it in practice and you experience how a single line of requirements simultaneously shakes the schedule, development cost, contract scope, maintenance, and liability.

Code is the same. Building a program and operating it for a year are different problems. With time, user behavior, accumulated data, changing requirements, external service changes, outages, and the technical debt of the original design all come back around. This is exactly the point Jobs was stressing: not making a judgment and leaving, but owning the results for a long time and gaining “scar tissue” from your own mistakes — only then does experience become three-dimensional.

So the essence of real-world experience is not simply field exposure.

It is a feedback loop with responsibility attached.

Repeat that process and knowledge turns into judgment.

And in the AI era, this difference will likely matter even more. Generative AI is already lowering the production cost of knowledge work — writing, summarizing, translation, coding. In experimental studies compiled by the OECD, generative AI cut completion time for certain professional writing tasks by an average of 40% while improving quality by 18%, and in a customer-support study, hourly productivity rose about 14% on average after AI adoption. The effects were largest for less experienced workers.

The ILO estimated in 2025 that roughly one in four workers worldwide is in an occupation with some exposure to generative AI. But it judged that most occupations are more likely to see their task composition change than to disappear outright. The OECD likewise notes that as AI takes over some repetitive, time-consuming tasks, humans may concentrate on higher-value work.

In other words, distinguishing people by “what they know” alone will get harder and harder.

Conversely, the OECD expects competencies like problem-solving, creativity, innovation, data interpretation, management, and human judgment to remain important in the AI era. In practice, most workers who use AI will not be specialists developing AI models, but people who must apply AI within their own work and exercise judgment over it.

So the question we ask about talent must change too.

It cannot end at what someone studied.

The OECD, too, has consistently recommended skills-based hiring, on-the-job training and apprenticeships, and stronger links between industry and curricula as the direction for reducing Korea's education-labor market mismatch. In other words, “look at real ability over credentials” is not just an entrepreneur's sentiment — it is a policy argument tied to structural problems in the Korean labor market.

If AI is making knowledge ever cheaper, Korea's talent system also needs to move from a system that selects people by how much they know to one that discovers people by how deeply they have experienced reality, how well they judge, and how much responsibility they can carry.