Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Your job isn’t safe because it’s physical

AI risk depends on how often work is repeated, not whether the job is physical or white collar.

[Photo: Getty Images]

Profile photo of Steve Chew

Co-Founder & Chief Strategy Officer · BuildOps

A commercial electrician’s week holds two very different types of work. Monday is a new build: conduit runs, panel schedules, a set of drawings, a sequence the crew has done a few hundred times. Tuesday is a service call in the basement of an aging office building where a rooftop unit keeps tripping, with no drawings, no signal on his phone, and a building engineer who describes the problem as “it makes a noise.”

Same man. Same license. Same tools in the same van.

Monday’s work holds a pattern that AI could learn. Tuesday’s job depends on what no one has written down.

I have spent the last eight years building software for commercial contractors, which means I spend most of my week looking at how this work gets done across thousands of companies. The thing I keep running into is that almost everyone is sorting jobs on the wrong dimension.

We ask whether work is physical or knowledge based, if it’s blue collar or white collar, does it require hands or head. This is the frame behind every AI exposure ranking I have seen, and it produces a comforting answer for the trades and an ominous one for everybody with a desk.

But it’s the wrong cut. The question that predicts AI exposure is much less flattering: how often does the same situation come around again?

Repetition drives AI exposure

Ignore the collar, and here’s what the exposure research shows.

Microsoft Research scored 785 occupations on how much of the work generative AI can genuinely touch. Interpreter and translator is the occupation that tops the list. Historians and writers are right behind it. These are the credentialed, thinking-for-a-living jobs that for decades we believed could never be replaced by technology. They lead the list because the work runs on patterns that recur constantly.

Meanwhile, software developers score higher for AI exposure than bookkeeping clerks and dispatchers—the jobs everyone assumed would go first. Software developers are exposed because most code has been written before, in public, at enormous volume.

Roofers are at the least risk, ranked at 770 out of 785, because their work changes with every job. A roofer makes decisions based on what turns up once the old roof comes off, like rotted decking or a leak nobody knew about.

The collar doesn’t matter in this ordering. What matters is how much of the work repeats.

SAME TRADE, TWO REALITIES

This is why the same trade splits down the middle. New construction is sequenced, specced, and governed by best practices that have been stable for decades. It is the most patterned work in the built environment, and it will absorb AI faster than anyone in the industry currently expects.

Service work is the opposite. A technician walks up to a system nobody documented, installed by a company that no longer exists, and diagnoses the problem from the sound it makes and years of having heard that sound before.

One is a pattern. The other is a judgment call in a crawlspace.

SORT BY WORK, NOT ROLE

Most AI plans treat a trade as one thing, so they overestimate what AI can do on the service side, and underestimate what it can do on the build side.

Almost nobody holds a job that is entirely one thing. The dispatcher who everybody has written off spends part of the day doing genuinely repetitive work and part of it deciding which of two angry customers gets the only available technician at 4 p.m. on a Friday—a decision that requires knowing that one of them is three months from renewing a service agreement.

If you sort your organization by role, you will cut the wrong people and protect the wrong work. You will cut people whose judgment you cannot replace, and you will protect work that an AI model quietly absorbed last quarter.

Organizations should sort by the work instead. Take any task on your team’s plate and ask: how many times has this exact situation come up before, and did the right answer ever change? High repetition and stable answers mean that AI is coming—and soon—regardless of the job title. Low repetition, or answers that shift with context nobody wrote down, means you are looking at something durable.

ENGINEERS FELT IT FIRST

I run a software company, so my engineers were the first to feel this.

A year ago, an engineer who spent the day writing clean, efficient code was doing exactly what we asked. Today, if that’s how he spends his whole day, something has gone wrong. Writing code turned out to be the most patterned thing we do—the most examples, the most precedent, the least novelty per hour.

What’s left is the hard part: framing the problem, deciding what is worth building, explaining a tradeoff to somebody who disagrees, and directing a set of agents that behave like fast, capable, wildly overconfident junior employees.

I will say the thing that doesn’t go over well at software conferences: being a tradesperson is a safer bet right now than being a software developer. Not because software matters less but because the developer’s day is far more patterned than the service technician’s, and pattern is what is commoditized by AI.

MAKE TWO LISTS

Take one team and split its work into two lists: situations that recur and situations that don’t.

The first list is your automation roadmap, and it is probably longer and more senior than you expect. The second list is your actual competitive position, and it is probably held by people you have never thought of as strategic—the estimator who knows which general contractor always disputes the final invoice, the service tech who can hear a bad compressor through a wall.

Then tell your people which list they’re on, and what you expect them to do about it. That conversation is uncomfortable and most leaders are avoiding it, which is why their teams have quietly concluded the worst.

The work that survives this isn’t the work done with your hands. It’s the work that is never quite the same twice.

Steve Chew is cofounder and chief strategy officer of BuildOps.