80% comes quickly; the remaining 20% takes 10 years

At the center is the record of a small parts maker with about 80 employees that took on a humanoid as an intern and put it on its washing process for 6 months. What matters in practice is that the target process was chosen not for its added value but because it was the station where people endlessly carried boxes of under 15 kg holding six jigs, causing musculoskeletal disorders. As a baseline for expectations, it presents a long-tail diagnosis (30–50% of a skilled worker's level in one Chinese company's factory trial, a goal of 80% in 3–4 years, and another 10–20 years for the remaining 20%) and sets beside it the case of a 2017 declaration of full automation that reached only 3.7% of its weekly target. The scene in which a soccer robot stopped in front of a red carpet because it had learned the ball only as "a red circle" is exactly how an inspection model breaks down when it has learned proxy features such as background or color rather than the defect itself. The answer from a worker with 40 years of experience, that 70–80% of their skill can be carried over into documents and filmed records, becomes a realistic target for archiving know-how.

This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.