The Wrong Battle
Girish Bhatia of ConstructMind.AI at Project Controls Expo in October of 2024.
We keep asking the wrong question about AI in construction.
AI versus human. Who wins. Who gets replaced. Who loses their edge.
But when you talk to Girish Bhatia, founder of ConstructMind.AI, he will tell you that this is fundamentally the wrong belief from the start. Yet, up until now, we've seemed to build our entire AI adoption strategy around that core belief.
But the thing is, it was never truly about us versus them to begin with.
It's a Sequence, Not a Fight
To be clear, the real architecture isn't even a battle at all. It's an order of operations.
Human. AI. Human.
A human carries the institutional knowledge, the 10, 20 or even 50 years of hard-won judgment baked into how construction teams actually build. Then AI does what it's good at, things like calculation, formatting and heavy lifting a person could do that would take ten times longer. And then finally, a human validates the answer before it becomes a decision.
Pull any one of those three out and the whole equation breaks down.
Without that first human, the AI is computing with garbage that has no context, no judgment and no idea of what "good" looks like on a job. Take out the last human, and you're shipping unvalidated outputs straight into your literal forecast, which is never safe. Take out the AI, and you're back to doing in three weeks what should take an afternoon (or less).
You see, the sequence becomes the product. Not the model.
The key is noticing where both humans actually live. They're not in the back office simply reading a dashboard. They're on the front lines, like the superintendent who knows why that pour slipped or the foreman who can look at an AI-generated progress update and tell you in two seconds it's wrong.
The model in the middle is the easy part. But the judgment on both ends is the whole game.
What Kills Adoption in AI
Even in these early phases of the hype cycle, most AI rollouts in construction aren't failing because the tech is bad or not ready.
They're failing because folks are waging war instead of designing a workflow.
Be honest, your first thought in buying an AI tool is the idea of replacing people, not of bracketing your people around it. And while the buying committee saw a demo of immense productivity gains, the folks who'd actually use the thing were never even asked what they thought. Instead, they watched as a tool was dropped on top of them and secretly plotted not to adopt the very thing that was brought in to make them obsolete.
Of course they did. You and I would too.
This is the part that Girish has realized we keep getting backwards. Adoption isn't a training problem at all. It's a design problem. When the design treats people as the thing to be removed, the people remove themselves from the rollout. Resistance no longer becomes about stubbornness, rather it's a rational response to a tool that picked a fight with them.
But if you build the loop so that your people are the bookends and not the casualty, suddenly that resistance has nowhere to stand.
We Build Tech for the Wrong End of the Job
Unfortunately, Girish is also quick to point out that’s not the end of the story.
When we buy software, we optimize for the output. The report. The dashboard. The thing the executive wants to see on Monday morning. So the typical tool seems to always be designed in a way that makes reporting easy. Especially for that handful of people who consume reports.
And nobody optimizes for input.
The problem is, the input is the entire job. Somebody in the field has to feed the thing. So as we've spent thirty years building tools that make the reporter's life easier, we’ve seemed to make the field's life much harder. More clicks, more forms, more double-entry and more "just log it in the system" on top of the work they already do. Then we act surprised when adoption craters.
If you want the human-AI-human loop to actually run, focus first on building for the front of it. Make the input effortless, or the loop is surely going to starve. A model, AI or not, will only be as smart as the field data feeding it, and field data will only show up clean when capturing it costs the field nothing.
It’s the same mistake we've been making since we traded paper for Excel. Taking an old process, making it digital and never once asking whether the old process was right to begin with. That really wasn’t innovation at all. That was digitizing. If we don’t get this workflow right, AI is going to expose every place we did that. You simply cannot bolt intelligence onto a workflow that was broken from the beginning.
Which as Girish puts it, is exactly why field-focused tech is no longer a nice-to-have. It's the only version of a future that works.
It’s Already Happening
If you think you still have time to get out in front of the issue, you may need to think again. While leadership teams debate whether to let AI into their construction business, their people already did.
ChatGPT. Claude. Gemini. There’s a guarantee one, if not all of these, are running in your offices, on your projects and with your data off the books. Girish calls it “building a dragon” where every query is silently making someone else's model smarter while your hard-won lessons learned leak out the side door.
In today’s landscape, we no longer get to decide whether AI is in our organizations. That ship sailed long ago. We only get to decide whether the human-AI-human loop runs within our own walls or outside of them, whether in an enterprise instance we control (learning from only our projects), or in a public tool that pockets our knowledge and hands us little in return.
And while the vendors are out there pushing us to buy more, whether it be another platform, another point solution or simply another feature. Girish's fix is the opposite. If you've got fifty tools, get to three. Fewer tools that actually talk to each other beats a stack of disconnected, best-in-class point solutions every time. The loop can't run across forty silos that don't share data, and this dragon you want to feed needs one body, not forty.
The Battle was Over Before it Began
It’s time to quit squinting at the future while trying to predict who is going to win. It was never really about that in the first place.
If you’re stuck on the mindset of AI versus human, you’re going to be left behind. It was always human, then AI, then human. Not us versus them. And the organizations that thrive won’t be the ones that replace humans with AI.
They'll be the ones that design better loops. Human. AI. Human.
That’s not the future of construction. It’s already here.
Construction is cool, tell your friends!