I'm on the wrong side of your chart — software delivery, the 75% column. So take this as a field report from the exposed side.
The number is real but it measures the wrong thing. It's how much of the task surface AI touches, not how much of the job it does. Last year I ran sixty stakeholder interviews on a delivery project and had AI aggregate them. It was genuinely useful and it saved me hours. It also invented requirements nobody asked for, quietly turned two groups who disagreed into one group who agreed, and dropped every exception case because exceptions are the part people mention as an aside.
Every one of those failures is the same failure: judgment, sequencing, relationships. The exact three things you say protect construction.
So the honest version might be that construction isn't safe and software isn't doomed. It's that the residual is the same everywhere — it's just a bigger share of the job on your side. We're both still doing the part the model can't do. You just have more of it.
Which makes your "build the AI-native team inside the judgment-heavy industry" point stronger than the report suggests, not weaker.
The financial side of construction is a perfect example of this. Job costing, cost to complete, WIP reporting — these aren't tasks that AI is replacing anytime soon, because the judgment required to interpret what the numbers actually mean in context is entirely human. A PM who understands why their cost to complete shifted and what to do about it is doing exactly the kind of high-judgment work you're describing. The ones who just report the number without understanding it are the ones who are genuinely at risk — not from AI, but from being irrelevant in a room full of people who do understand it. Great framing on the exposure data.
I think the real opportunity (and I know that you hit it at the end of this post) is the layering of AI on top of physical businesses. I think that is perhaps the biggest wealth unlock of this decade. Companies that build things, companies that provide things, companies that do things for others: any of those, when layered with AI, have incredible potential. Great posts and it was wonderful to hear your take on what you are seeing, literally boots on the ground.
I'm on the wrong side of your chart — software delivery, the 75% column. So take this as a field report from the exposed side.
The number is real but it measures the wrong thing. It's how much of the task surface AI touches, not how much of the job it does. Last year I ran sixty stakeholder interviews on a delivery project and had AI aggregate them. It was genuinely useful and it saved me hours. It also invented requirements nobody asked for, quietly turned two groups who disagreed into one group who agreed, and dropped every exception case because exceptions are the part people mention as an aside.
Every one of those failures is the same failure: judgment, sequencing, relationships. The exact three things you say protect construction.
So the honest version might be that construction isn't safe and software isn't doomed. It's that the residual is the same everywhere — it's just a bigger share of the job on your side. We're both still doing the part the model can't do. You just have more of it.
Which makes your "build the AI-native team inside the judgment-heavy industry" point stronger than the report suggests, not weaker.
If you are reading this at your desk right now, you need to think about this article hard.
The financial side of construction is a perfect example of this. Job costing, cost to complete, WIP reporting — these aren't tasks that AI is replacing anytime soon, because the judgment required to interpret what the numbers actually mean in context is entirely human. A PM who understands why their cost to complete shifted and what to do about it is doing exactly the kind of high-judgment work you're describing. The ones who just report the number without understanding it are the ones who are genuinely at risk — not from AI, but from being irrelevant in a room full of people who do understand it. Great framing on the exposure data.
I think the real opportunity (and I know that you hit it at the end of this post) is the layering of AI on top of physical businesses. I think that is perhaps the biggest wealth unlock of this decade. Companies that build things, companies that provide things, companies that do things for others: any of those, when layered with AI, have incredible potential. Great posts and it was wonderful to hear your take on what you are seeing, literally boots on the ground.