“Please jot down some notes for yourself identifying at least three ‘key events’ in your career: things that made a difference in the way you manage now.”
In the 1980s, researchers at the Center for Creative Leadership (CCL) put some version of that request to 191 executives from six major corporations, five of them in the Fortune 50. Then two questions. What happened? What did you learn from it?
The answers came back as 616 stories and 1,547 lessons. One manager had been made production superintendent of the division’s largest plant, where there had been “threats, bodily injury, sabotage” and “two hundred outstanding grievances against us.” Another moved overseas, from marketing to manufacturing, without speaking the language. “I was in charge of 7 plants in 7 countries,” that executive recalled.
American business was spending over $40 billion a year to “educate” more than 8 million “students,” the team noted. Formal courses, the thing companies were spending billions on, made up 38 of the 616 stories.
Seventy, twenty, ten
Most of what these executives said changed them happened on the job and through other people, not in class. 70-20-10 is a later, rough summary of those memories, not a measured split.
In their 1996 Career Architect Planner, Michael Lombardo, one of the original researchers, and Robert Eichinger wrote that lessons learned by successful managers are “roughly” 70 percent from tough jobs, 20 percent from people, mostly the boss, and 10 percent from courses and reading, as a 2014 ATD article by Andrew Jefferson and Roy Pollock quotes them. CCL’s website now calls the rule “a research-based, time-tested, classic guideline.”
You can hold that shape against the 1987 technical report, which counts every category. The five kinds of assignment (start-ups, turnarounds, task forces, jumps in scope, line-to-staff moves) and the five hardships (mistakes, demotions, breaking a rut, problem subordinates, personal trauma) add up to 401 of the 616 stories, about 65 percent. Bosses and other people account for 111, about 18 percent. Courses, 38, about 6 percent. The remaining 66, early jobs, first supervisory jobs and purely personal events, fit none of the three.
So the ratio wasn’t pulled from the air. But look at what was counted: the stories successful people chose to tell when asked for turning points. A two-year turnaround and a two-week course each count as one. Nothing in the data says how much anyone learned from each.
A fact in search of evidence
The report’s authors said so themselves. “All of these data were self-report,” they wrote, and they had “no way of knowing for sure that the events happened as they were described, or that the executives actually learned what they claimed.” The 191 were picked as high-potential or successful; 200 more executives in two companies, not seen as high-potentials, put their events and lessons in a significantly different order.
Jefferson and Pollock point out that the phrase 70-20-10 never appears in the 1988 book. The model, they conclude, is “neither a scientific fact nor a recipe for how best to develop people.” In 2018 Alan Clardy reviewed five research traditions that seem to agree that 70 percent or more of work-based learning is informal. He judged the evidence for that “70% rule” weak and wrote that it “needs to be set aside.”
The same year, a study of middle managers in the Australian public sector by Samantha Johnson, Deborah Blackman and Fiona Buick found the framework not delivering. One reason was an overconfident assumption that unstructured experience automatically builds capability.
What the stories do show
Drop the percentages and the report still maps which experiences taught what. Assignments taught the business, toughness and how to win cooperation from people with no reason to give it. Hardships taught people about themselves. Values came mostly from other people. And the assignments, the authors noted, “are the largest block of events that an organization can directly control.”
Courses mattered when the timing was right. One manager took a course in organizational design while on a task force redesigning the corporation.
Imagine a manager told by HR that 70 percent of a promising engineer’s development should come “from the job.” The engineer keeps the same job for another year, now labelled a stretch. Nothing in it is new, so there is little to learn. A real jump in scope, a senior colleague to talk it through with, and a short course timed to the first month would each have had work to do.
When the assistant does the hard part
The executives’ lessons came from problems they had to work through, “building new skills on the run,” in the report’s phrase. AI assistants now take some of those problems away.
In a randomized experiment posted as a preprint in January 2026, Judy Hanwen Shen and Alex Tamkin, working with Anthropic, had 52 programmers new to a Python library called Trio complete two tasks with it; half could use an AI assistant. Afterward, without AI, the assistant group averaged 4.15 points lower on a 27-point quiz about the library, with the widest gap on debugging, and they weren’t significantly faster. The median participant without AI hit three errors along the way; with AI, one. The authors think working through those errors helped the others learn. Within the AI group, the small clusters who asked conceptual questions or wanted explanations with the code averaged 65 to 86 percent on the quiz; those who leaned on it to write or debug averaged 24 to 39.
It was one hour, one library and a quiz, with no real job riding on it, and the authors call it a first step. Whether a year-long stretch assignment teaches less when an assistant handles the parts that used to go wrong, this study didn’t measure.