AI may automate the pipeline that creates experts before it automates experts themselves

imjustnewatai · x · 2026-07-27

i watched all 3 dario interviews, checked what actually happened months later, then compared it with the 1,171-career map we built.

The core argument is that AI may automate the process that creates experts before it fully automates the experts themselves.

Anthropic’s data is cited as showing humans still make about 70% of planning decisions, while Claude handles about 80% of execution decisions.
Expert users reportedly achieve verified success more than twice as often as novices.
The catch: execution work is also how beginners become experts.

The thread argues that if agents absorb drafting, debugging, spreadsheet reconciliation, contract review, and edge-case learning, companies may need fewer juniors now and society may produce fewer experts later.

It frames Dario Amodei’s “90% hump” as a near-term state where humans supervise the remaining 10% of highly leveraged work.

It also says the final step change may come once agents get memory, tools, computer access, feedback, verification, and permission, turning “a worker using AI” into “one expert supervising an AI organization.”

The broader claim: many careers collapse into a handful of reusable loops, and the first labor shock may look less like mass layoffs of experts and more like companies quietly stopping the hiring pipeline that creates them.

Original post →

AI may automate the pipeline that creates experts before it automates experts themselves

More from AGI Musings