What I told the project manager who showed me the article on his phone and asked if AI is going to kill us all
It was a little after seven in the morning on a Thursday, and I was in a Tempe office waiting on a superintendent to finish a call to start our meeting. A project manager who was part of the meeting was scrolling his phone with the specific frown guys our age reserve for either bad sports news or something from the news app. He turned his phone to face me. “Since you do the AI stuff, is this true? Are we supposed to be worried about this?”
The headline on his phone was about the head of a major AI company warning that AI agents could become dangerous enough within the year that the industry needed serious new safeguards. A few days before that, a researcher had quit that same company, publicly, and estimated the odds of AI causing human extinction within a decade at somewhere around one in ten. That same week, a couple of the loudest voices in AI hardware and open source software called that number nonsense, in public.
I did not have a clean answer for him. I still don’t. But I have spent twenty years watching how construction people process risk, and I told him what I’ve told others on the topic. I have no idea if the real number is one percent or ten. What’s more, nobody arguing about it in public right now actually knows either. What I do know is what I am going to do about it Monday morning, and that part is not complicated.
“You do not have to resolve whether AI ends the world to decide what to do about it in your own company. You only have to handle the ten percent that is actually yours to handle.”
Two Guys Who Build the Thing, Disagree
Here is what actually happened this week, stripped of the noise. The CEO of one of the top AI labs said publicly that a swarm of AI agents working together might be capable enough within six months to a year to cause real damage if companies do not slow down and build in more safeguards first. Days earlier, a researcher who had just resigned from that same company said he thought there was roughly a ten percent chance AI causes human extinction within the next decade, and accused the major labs of racing toward more powerful systems faster than they can control them.
Other people who build this technology for a living think that is nonsense. The head of the world’s dominant AI chipmaker flatly dismissed the extinction talk as nonsense, in public, that same week. Another AI company’s CEO compared asking a researcher to grade the odds of human extinction to asking your air conditioning guy to weigh in on climate change. And a formal report written with input from more than a hundred independent AI experts landed right in the middle, describing the risk’s likelihood, nature, and timing as “unusually ambiguous.”
That is not a fringe argument happening somewhere on the internet. That is the people closest to the technology, disagreeing with each other, in public, in the same week. If the people building the thing cannot agree on how dangerous it is, I am fairly sure that it’s unrealistic to expect a general contractor in Arizona to have a firm opinion by Friday.
The Number I Actually Carry Around
Here is what I have watched for over twenty years. The people who run good companies do not wait around for certainty before they act, and they do not throw out a real number just because it makes them uncomfortable. A ten percent chance of something catastrophic is not nothing. It is also not a coin flip. It is the kind of number a good operator sits with, the way you would sit with a sub who has a shaky safety record but is the only one who can hit your schedule, or an insurance renewal that just jumped because the market got nervous.
So I sit with it. I do not lose sleep over superintelligence. I am definitely not smart enough to referee that argument, and neither is almost anyone writing about it. What I do think about is a much smaller, much more knowable ten percent. The odds that something in my own company, or one of my clients’ companies, is quietly running an AI tool nobody approved, touching data nobody is tracking, making a decision nobody could explain if a client asked. That risk is not theoretical. That risk is sitting on a laptop somewhere in Arizona right now.
The Race With China Doesn’t Change What You Do Monday
The other half of what my PM’s phone was telling him is real too. The United States and China are genuinely racing on AI, on chips, on export controls, on which country’s models end up running the world’s software. Recent estimates put U.S. AI investment north of seven hundred billion dollars in a single year, against roughly two hundred billion dollars China has spent building its own AI industry over the past decade. Members of Congress have called it an indefinite race. Executives on both sides call it existential.
None of that changes what is sitting on your file server in Tempe. If you have been in construction in Arizona for more than about fifteen minutes, you already know the difference between a real problem and a headline problem. The chip race is going to decide who has the fastest models in five years. It is not going to decide whether your submittal log is governed next month, whether your estimator’s AI copilot is trained on last year’s pricing, or whether anyone in your office knows what data an AI tool can see. That part is entirely up to you, regardless of who wins the bigger race.
We Have Been Here Before
I want to be careful here, because I am not telling you to relax. But I am telling you this pattern is old.
When the first railroads opened in the 1830s, doctors publicly worried that the human body could not survive speeds over thirty miles an hour, that organs would shift or the sheer velocity would tear the soul loose from the body. In 1955, President Eisenhower dismissed fears that automation would gut American jobs, calling them groundless and noting that the same warnings had been showing up for roughly a hundred and fifty years. Every decade since has produced its own version. The New York Times ran a story warning a robot was after your job in the 1970s. The entire country spent 1999 bracing for Y2K to black out the power grid. None of those disasters happened, not because the fear was crazy, but because a lot of unglamorous, unphotographed work got done in the background to make sure it did not.
That history does not prove AI risk is zero. It proves something more useful for a business owner. Every generation gets a technology that looks like it might end everything, and every generation has to figure out how much of the fear is real and how much is theater, while still doing the boring work that actually keeps them safe. That is the whole story in a single sentence.
Meanwhile, the boring work in construction is not getting done fast enough. AGC and Sage’s 2026 outlook found 61 percent of construction firms are now using AI or planning to increase their investment in it, up from 44 percent the year before. FMI has put the industry’s annual losses from rework tied to bad or miscommunicated project data at 31 billion dollars a year. Adoption is accelerating. Governance is not keeping pace with it. That gap is where the real risk lives, for you, this year, regardless of how the extinction debate turns out.
THE GOVERNANCE GUT CHECK
Who at your firm can turn on a new AI tool this week without IT or ownership signing off?
If an AI tool made a bad call on a submittal, an RFI, or a schedule this month, would anyone know which tool, which data, or how to undo it?
Where does your most sensitive project and client data actually live, and who has access to it right now?
Is there one written document at your firm that says what AI is, and is not, allowed to touch?
If you cannot answer all four of those in under a minute, that is your real ten percent problem. Not the one in the headlines.
Your Next Step
Nine times out of ten, the fix here is not a new policy binder or a company wide AI ban. It is smaller than that, on purpose. Before your next OAC meeting, spend twenty minutes writing down every AI tool anyone on your team is currently using, official or not, and what project or client data each one can see. Estimating copilots. Chatbots drafting emails. A super running jobsite photos through an app nobody approved. Write it all down in one place.
You cannot govern what you cannot see. Most Arizona firms cannot see it yet. Fixing that list this month does more for your actual risk than any opinion I could offer you about whether AI ends the world in ten years.
I never did give my PM a real answer on the extinction question. I told him what I am telling you. Nobody arguing about it in public actually knows the number, and I am not going to pretend I do either. What I do know is that nobody in that story, not the CEOs, not the researchers, not the guy sitting in a Tempe office, was trying to hack anybody, lie to anybody, or miss the target. They are all just trying to see a genuinely new kind of risk clearly, at the same time everyone else is.
You do not have to solve that one. You have to solve yours.
The ten percent that keeps me up is not the one in the news. It is the one sitting ungoverned on a laptop in your office right now.
For over 20 years, Dimensions has been the trusted IT partner for Arizona’s architecture, engineering, and construction industry. We help AEC firms communicate better, collaborate smarter, and actually use the technology they’ve invested in. Because in construction, the tools only work if your team does.
IT Built For Builders.
