What Judgment Work Costs

For the last two years, organizations treated the human layer around AI as friction. Time spent validating outputs looked like inefficiency. Someone questioning a recommendation looked like hesitation. It was work, but it was rarely counted as work, so it never showed up on a business case.

That layer has a price now. Several organizations have paid it.

In July of 2025, Commonwealth Bank of Australia announced it would cut forty-five customer service roles after introducing a voice bot the bank said had reduced call volumes by two thousand a week. The Finance Sector Union disputed the figure, saying volumes were rising and the bank was covering the gap with overtime and by pulling team leaders onto the phones. By late August the bank had reversed the decision entirely, telling employees its initial assessment had not adequately considered all relevant business considerations and that the roles were not redundant after all.

The bank had modeled what the system could absorb. It had not modeled what was left over.

Ford paid a larger version of the same bill. Bloomberg reported in June that the company spent three years hiring back three hundred and fifty veteran engineers, many of them former employees, to address quality problems that had cost it billions. Charles Poon, Ford’s vice president of vehicle hardware engineering, said the company had assumed that introducing AI and feeding it existing design requirements would be enough to produce a quality product. It wasn’t. The automated systems were not delivering the results the company expected.

What Ford discovered is more specific than a failed deployment, and more useful.

The AI tools underperformed in part because many of the company’s most experienced engineers had already left before their knowledge could be captured. The systems were trained without the people who knew what to look for. The returning engineers are now rebuilding those tools and running design reviews to catch failure points before parts reach the plant floor. Ford took the top spot among mainstream brands in this year’s JD Power Initial Quality Survey, its first in sixteen years.

None of this was unforeseeable. In October of 2025, Forrester predicted in its Future of Work report that more than half of AI-attributed layoffs would be quietly reversed as organizations discovered the operational difficulty of replacing people prematurely. That was a forecast at the time. It is now a pattern with names attached.

The judgment layer was always there. It sat distributed inside jobs, unnamed, absorbed into roles that appeared to be about something else. Answering a call. Reviewing a design. When those roles were removed, the judgment went with them, and the work it had quietly been doing became visible for the first time by its absence.

Buying it back is expensive, but it is available. That should be uncomfortable rather than reassuring.

It is available because there is currently a supply of people who developed that judgment somewhere else, under conditions that no longer exist. Ford’s returning engineers spent decades seeing enough cases to know which one doesn’t fit.

Every organization solving this problem right now is buying judgment rather than building it. That works exactly as long as the supply holds.

You can buy an experienced engineer. You cannot buy the experience itself.