AI is rapidly changing the value of procedural knowledge—the ability to follow established steps, workflows and instructions to accomplish a task. For decades, businesses have rewarded people for becoming efficient at repeatable processes: writing standard reports, analyzing routine data, producing boilerplate code or following established operating procedures. As AI agents become increasingly capable of executing these workflows, the economic value of simply knowing how to perform a procedure is likely to decline.
This does not mean that human expertise disappears. Instead, AI creates a sharper distinction between procedural work and judgment-based work. AI is particularly effective when the objective is clear and the process can be described or learned from examples. But ambiguous situations, unusual problems and decisions involving competing objectives still require context and judgment. Research on AI-assisted knowledge work similarly suggests that AI capabilities are uneven: systems can dramatically accelerate some tasks while making complex verification and oversight more important.
That creates an important challenge for education and career development. Traditional professional training often works by teaching people procedures first and gradually giving them more responsibility. If AI can perform many entry-level procedures immediately, workers may have fewer opportunities to acquire the experience that eventually produces expert judgment. This connects with a broader concern in software engineering, for example, where AI can perform junior-level coding tasks while potentially removing some of the work through which junior developers learn the fundamentals.
The article's larger idea is that the AI era could produce a “procedure economy” in which following instructions becomes increasingly cheap while deciding what should be done becomes increasingly valuable. The competitive advantage may shift toward problem framing, original thinking, experimentation, verification and knowing when established procedures no longer apply. This is consistent with broader research suggesting that organizations will need to redesign work around AI rather than simply automate existing tasks.
