A Well-Overdue Construction Overhaul

When you think of AI, what comes to mind first?

If you're in the trades, it's probably one of these:

- "AI can't do my job." - "It's cool, but I don't know how it affects me." - "What am I supposed to do if it takes my job?"

Usually in that order. I went through all three in the summer of 2023 when I first played with ChatGPT 3. I was deep in architectural visualization at the time and I did what a lot of people do: I actively chose to stop paying attention to it. I leaned harder into ArchViz, then harder into Midjourney and DALL-E when those started getting interesting. I built a small Etsy store dropshipping AI-designed products, then wrote a program that would generate designs on request from a prompt. I was capitalizing on novelty more than understanding anything.

What changed things was going back and actually learning what these tools are.

What "AI" actually means

The term AI is about as broad as "the internet." When someone says AI, they might mean a recommendation engine, a chatbot, an image generator, a satellite navigation system, or a protein folding model. These are not the same thing.

A cleaner way to think about it:

LLMs (Large Language Models) are the text engines — ChatGPT, Claude, Gemini. They process language and generate responses based on patterns learned from enormous amounts of text.

Multi-Modal Models are the current generation. They handle text, images, audio, and actions in one system. They don't just answer questions — they can plan and execute tasks across different types of input.

Inference is the part most people skip, and it's the most valuable for businesses. It's using your specific data to ground these models in your actual situation. Not generic answers — answers shaped by your workflows, your numbers, your company's specific patterns.

That last piece is what's massively underused in construction.

What construction actually needs

Construction firms have data everywhere. Inbound emails with specs and change orders. Supplier quotes arriving in every format imaginable. Project notes scattered across six different apps. Foreman updates in a group text.

None of it talks to each other.

What inference makes possible is a system that reads all of that — parses it, derives meaning, and routes it to wherever it needs to go — without someone manually re-entering it somewhere else. The model doesn't need to understand construction in the abstract. It needs to understand your construction company well enough to make useful decisions about information as it arrives.

I've spent the last two years building exactly this kind of infrastructure. Not generic AI demos. Not chatbots. Systems that parse inbound chaos — emails, order forms, carrier notifications — extract structured data, and push it into workflows the rest of the business can act on.

The transition I didn't see coming

I attribute a significant chunk of my career to a moment when I stopped thinking of AI as competition and started thinking of it as the thing I'd been trying to build manually for years.

At my last firm, I built M365 automation that parsed meeting transcripts, extracted commitments and dates, and logged them without anyone touching a keyboard. I wrote scripts that pulled data from every inbound channel and categorized it automatically. I was doing AI implementation before I called it that.

The label "AI Implementation Strategist" landed on me later, but the work had been accumulating for a while. Construction is behind on this — not because the people in it aren't sharp, but because most of the tools being pitched at them were designed by people who've never spent a Tuesday on a job site.

That's the gap this whole thing is built around.