The word of the year for 2025 was slop. That tells you everything you need to know about where we are with AI right now.
I got a text a few weeks ago from a doctor's office I visited months ago. No details, no itemization, just "you have a balance due." I remember paying. I don't remember any additional charges. So I'm sitting in traffic, calling an 800 number, pressing 1 for this and 3 for that, clicking through fifteen menus just to earn the privilege of being put on hold. When I finally got a human, I had to argue about a charge I'm pretty sure I don't actually owe.
That is slop. Not the AI kind. The kind we've been living in for twenty years.
People keep telling me AI is going to make the world less human. I think they have it exactly backwards. The most human thing that could happen to my afternoon is an AI that picks up on the first ring, knows who I am, pulls up my chart, and either resolves the question or tells me straight. That's also one less person stuck on a headset all day getting yelled at by strangers about medical bills they didn't cause. That job is drudgery. That job should be automated.
AI that removes friction isn't the enemy of the human experience. It's the thing that gives us the human experience back.
The cold-email objection
I hear the opposite take constantly in construction: AI is going to kill the human part of this business. I get where it comes from. Relationships are everything in construction. You lose those, you lose the whole thing.
But look at how GCs are already inviting subs to bid today. Most of it is a BCC'd blast to every sub in a 2,000-mile radius. Generic subject line. Sometimes the wrong state on the project. Sometimes the sub's name isn't even on the email. Every sub I've ever talked to has the same reaction: annoyed at best, ignored at worst.
Is that a human experience?
Run the thought experiment. You're a sub. Two bid invites hit your inbox on the same Wednesday. One is the BCC blast — your name spelled wrong, no project address, no scope, somehow you're on a list for a job in Arizona when you only work in Missouri. The other has your name on it, addressed to you personally, with quantities pulled, drawings filtered to your trade, site location, access constraints, and an instant answer to the one question you were going to ask anyway. Which one are you responding to?
The phone call after — "hey, I sent you something, you want to bid this?" — that part matters. That part is sacred. Nobody is trying to automate that. But the work that comes before the phone call, the work of making the email worth opening, that's where AI earns its keep. Let the agents do the prep on both ends. Nobody wants to be the guy chasing a sub with ten follow-ups while the sub is working a 9-to-9 and just wants to get home.
The spreadsheet killed the relationship a long time ago
Once you see it that way for the outside of the business, you start to see where the real problem lives. It isn't in the emails between companies. It's in the work happening inside them.
I was listening to an interview with Jen, the CTO at Baker Construction. She told a story I haven't stopped thinking about. She was on a jobsite in Denver and watched a project engineer walk down from the ninth floor of a building, all the way to the conex on the ground, just to order a piece of material they needed on the pour. Then back up.
That's the shape of the modern construction day. A PE who got into this work because they love being out there spends half their life in logistics. They drive back to the office to cost-code an invoice. They leave the floor to chase an RFI. They turn into part-time IT because nobody gave them a real one — Baker has 400 to 600 projects going at once, and you can't put a tech person on every site. So the PEs adapt. They buy their own cradle points. They negotiate their own ISP contracts. They put Starlink on a personal credit card.
That is where the relationships actually die. Not when AI answers an email. When the foreman wants to walk the ninth floor with his PE, and the PE has to go do time entry instead.
The spreadsheet killed the human touch in construction a long time before anyone showed up with an LLM.
The backbone is about to get cheap
Jen had another line I keep chewing on. Construction, like most industries, spends most of its technology budget on what companies call shared services. ERP. HCM. Supply chain. Finance. Inventory. The backbone. And every business on earth runs those processes more or less the same way. You onboard someone. You move them. You offboard them. You buy a thing, you receive it, you pay for it. It's the same loop whether you're a concrete contractor, a software company, or an airline.
The companies selling that software figured out how to corner the commodity. They charge a lot of money to do the thing every business already does. And because most of the tech budget goes there, very little is left for the work that actually makes money — estimating, field coordination, the part of the business that differentiates you from the GC down the street.
Here's what's interesting about AI right now. It's going to flatten that commodity layer. The onboarding-moving-offboarding-purchasing-receiving-paying loop is exactly the kind of work agents are good at. Not because it's impressive. Because it isn't. It's repetitive, the rules are clear, and it generates slop when a human does it. It's the kind of work that makes a human feel like a cog.
When that layer gets cheap, two things happen. One, you suddenly have money to spend on the work that actually makes money. Two, you have attention. Your best people aren't buried in forms anymore. They're back on the floor. Back with the subs. Back with the work.
We could still build beautiful things
I watched a video about the River Thames a couple weeks ago. The guy was standing next to two lampposts. One was modern. It works. It lights up the sidewalk. It's fine. The other was from 1870, built into Joseph Bazalgette's embankment — ornate bases, bronze fish wrapped around the column, flourishes nobody asked for. Both were mass-produced. Both came out of a factory. The Victorians weren't doing handcraft on every lamp on the waterfront. They were doing industrial production, same as we are. They just decided the industrial production should be beautiful.
The punchline of the video was this. Bazalgette also built a pumping station to handle London's sewage. It's now a museum. People pay money to tour it. A sewage facility. Turned into a tourist attraction because it was built with ornamental exuberance, because the Victorians believed the people who worked in the sewers deserved a beautiful place to work. Compare that to a modern treatment plant. Or an office park. Or an apartment building. Or an AC unit stapled to the outside of a window.
We could still build like that. We just don't. We chose convenience. We chose margins. We chose the cheapest version every time, and over enough decades we forgot that a different choice was ever possible.
When people write about AI slop — the auto-generated art, the hollow music, the six-fingered hands — they frame it as a new problem. It isn't. The beige concrete box is slop. The stapled-on AC unit is slop. The phone tree is slop. The BCC'd cold email is slop. Slop is what you get when you optimize for convenience over care, and we've been doing it to the built environment for seventy years. AI didn't invent the disease. It just made the symptoms easier to see.
Which is why I think the word of the year being "slop" is a good sign. It means people are finally naming it. It means taste might be coming back.
AI doesn't have taste. Thank God.
Here's what I believe, and it's why I do this work. AI doesn't have taste. That's the thing it can't do. It can run a thousand takeoffs, but it can't tell you whether the number coming back makes sense. It can draft a thousand cold emails, but it can't tell you which sub is worth calling. It can generate a thousand scope sheets, but it can't tell you which scope gap is the one that'll bankrupt the project.
Taste is the thing humans keep. Judgment is the thing humans keep. The foreman's instinct for whether a pour is coming off right. The estimator's gut for whether a sub is lowballing. The architect's feel for whether the room wants a window on that wall or the next one. AI can do the drudgery around those decisions — the takeoff, the logistics, the data cleanup, the follow-up. But the decisions themselves, the taste, the care, the choice to build something worth looking at — that's still us.
That's the deal I'm excited about. AI eats the slop work. Humans get the soul work back. The 197 hours of phone tag per project. The nine-floor walks. The manual re-extraction of quantities from a PDF that came from a Revit model that already had them. All of that can go. And when it does, the people left standing are the ones who care whether the building is worth looking at in 150 years.
I'm the AI guy. I love this technology. I build with it every day. But the reason I'm excited about it has very little to do with the technology itself. It's that we've been living in a slop era for a long time, and AI is the first tool that's good enough at slop to maybe, finally, set us free from it. The thing that hollowed out construction wasn't a robot. It was the slow creep of administrative work into every corner of the job. The thing that'll put the soul back is whatever gets that work out of the way.
Call me a romantic. I think the 2030s could be construction's renaissance. I think the next generation of PEs could spend their days on the ninth floor instead of the trailer. I think the BCC'd cold email could become a memory. I think we could build things worth standing in again.
A de-IKEA-ification of the entire built world.
We just have to decide that's what we want.