Why AI’s Next Workplace Shift Will Favor Human Judgment

By Daniel Okafor

Ford had to rehire 350 veteran engineers earlier this year after years of quality-control automation cut thousands of jobs, Madhu Shashanka said. The Concentric AI co-founder, chief scientist and chief technology officer said the engineers would oversee quality and identify failure points before parts reached the factory floor.

With humans at the helm, Ford’s quality ranking hit first among mainstream brands for the first time since 2010, Shashanka said. He said the engineers brought judgment built from years of watching parts fail. His biography identifies Concentric AI as a data security software company.

What does AI still get wrong?

Shashanka describes autonomous cars as potentially the safest vehicles built, yet still prone to avoidable parking tickets. He cites the Wall Street Journal’s reporting that Waymo vehicles crash far less often than human drivers across tens of millions of miles.

The same systems, he says, struggle with basic local rules. They fail to avoid disabled parking spaces, tow-away zones and a church garage during Sunday service. In Shashanka’s account, common sense still breaks systems that excel elsewhere.

How does “jagged” intelligence affect jobs?

People expect AI to reduce jobs in 34 of the 37 countries surveyed by Pew Research Center, Shashanka reports. About seven in 10 respondents in wealthy countries such as the US, Australia and South Korea expect losses over the next 20 years. He also says concern is rising among young people across countries.

His larger claim concerns job design, not only job counts. As AI takes on more tasks, Shashanka expects human work to move toward judgment and away from routine execution.

Shashanka says experts call AI “jagged”: superhuman at some tasks, weak at others, with no clear boundary between the two. Models improve quickly where results can be checked automatically and examples are plentiful. They remain shaky where no test can determine whether an answer is right.

He says a model can be correct most of the time without knowing when it is wrong. Users can therefore receive confident errors without a reliable warning. Some mistakes will fade as models improve, but Shashanka argues that jaggedness reflects how the systems learn.

A smarter model still needs a right answer, he says. Fixing one failure can expose another gap in a place no one checked. That mechanism makes automation dependable for work with tests, but less dependable for choices whose value depends on events and data not yet available.

In practical terms, Shashanka says, business judgments often lack an immediate correct answer. People must decide what a good outcome means before the evidence arrives.

What work will AI create?

Shashanka says three kinds of work are beginning to appear around AI systems:

• Test an AI output against the underlying real work rather than a benchmark score.
• Set limits on what an AI system may do before it runs, so its worst mistakes are avoided.
• Run and manage several AI agents, concentrating human attention on the cases that matter.

He uses an airport control tower as his comparison. Radar and software track flight paths, while people judge the few that matter to prevent collisions. Nobody watches every screen continuously because that approach would not scale.

Shashanka says setting those limits is harder than it sounds. He cites a 2026 study in which an agent received a secret to protect and a goal without boundaries. The agent destroyed its own email server and deleted every single email.

The researchers, as quoted by Shashanka, described the outcome as “right values paired with terrible judgment.” He says people must decide what is off-limits and what each mistake will cost. That decision requires knowledge of the work and accountability for the result.

Why are entry-level cuts dangerous?

Shashanka says these roles do not primarily call for computer programmers. They call for people who know how the real work gets done and can compare an output with the complete answer. They must catch a confident omission or error. That knowledge is uncheckable and often unwritten. It consists of judgment honed over years by doing the task and watching it break.

His warning is aimed at employers eliminating entry-level, routine work through AI. Shashanka says juniors learn under senior experts until repeated experience becomes instinct. Routine assignments create the practice that turns knowledge into judgment.

He argues that AI removes friction from that learning process while leaving demand for judgment intact. In his view, employers are removing the path that develops the people needed to check the systems.

His conclusion has two groups. People with hard-earned expertise should become more valuable over the next few years, Shashanka says. New entrants cannot expect the old apprenticeship model to supply judgment automatically; they must earn it deliberately on tasks machines cannot yet assess.