79% of Construction Leaders Know AI Is the Risk. Almost None Are Doing Anything.

79% of construction leaders say not adopting AI will put them at a competitive disadvantage.

I believe them. The number comes from CMAA — the Construction Management Association of America — and it tracks with every conversation I've had with peer firm directors over the past year.

What I don't see is most of them doing anything about it.

I run a 15-person engineering consultancy. The pattern is identical across every peer conversation:

"Yeah, I know AI is important." "We tried ChatGPT for a few things." "We don't really have time to figure it out."

That's not adoption. That's awareness without action. And awareness without action is the most expensive position in the industry right now — because the firms that wait are paying the cost of the transition without getting any of the upside.

The misread

Most firm directors hear "adopt AI" and picture something they don't have.

They picture a transformation roadmap, an AI strategy committee, an enterprise software contract, a CTO they can't afford to hire. They picture six months of consultant fees before anything works. They picture a vendor demo with the word "synergies" in it.

So they file the question under "things to revisit when we have time" — and they never do.

This is the misread. AI adoption at a 15-person firm doesn't look anything like the McKinsey deck. It looks small, fast, and mostly invisible. One person tries one tool on one workflow. They see if it cuts time by a third. If yes, they keep going. If no, they try a different workflow.

That's the whole strategy.

What actual adoption looks like at a small firm

Four steps. Two weeks. One person.

1. Pick your top 3 time-consuming repeatable workflows

Not the most important workflows. Not the most strategic. The most repeated. The ones you do on every project, every client, every week.

For most engineering and construction firms, the list usually looks like:

• Proposals
• RFIs and clarification responses
• Meeting minutes and action capture
• Progress reports
• Site supervision summaries

Pick three. Don't spend a meeting on it. The list is obvious if you've been running the firm for more than a year.

2. Test one AI tool on one workflow for 2 weeks

One workflow. One tool. Two weeks. That's the experiment.

Pick the workflow where the tool has the highest chance of working — usually proposal drafting, because the structure repeats across bids and most of what you write is variations of the same technical narrative.

The tool doesn't matter much at this stage. ChatGPT, Claude, Copilot — pick one and run it. The point isn't to find the optimal tool; the point is to learn what changes when you have AI in the loop on a real workflow.

3. Measure: did it cut time by 30% or more?

Time is the only honest metric at this stage.

Not quality (you'll obsess over it and never decide). Not adoption rate (one person can't have an adoption rate). Not ROI (the spreadsheet won't move on a two-week sample). Just time.

If the workflow that took 8 hours now takes 5 — keep going. If it still takes 7 — try the next workflow with the same tool, or try a different tool on the same workflow. Don't quit on AI; quit on that specific combination.

4. If yes — standardize it. If no — try the next workflow.

This is where most firms fail. They test something, it works, and they don't standardize.

Standardization means: the prompt is documented, the workflow is repeatable, the second person can do it without asking the first person how. That's it. No platform required. A shared note will do.

Once one workflow is standardized, the next one is easier — because the firm has now seen the loop work and knows it's not magic.

Why "starting" beats "picking the best tool"

The firms that move first won't win because they picked the best AI tool. The tools converge. They're already converging. In 12 months the difference between Claude and ChatGPT on routine drafting work will be effectively zero, and the niche tools will all be cheap.

The firms that move first will win because they started while everyone else was still talking about it.

Two things compound when you start:

• Your prompts get better. Each iteration teaches you what to ask for and what context to include. By month six, your team's prompts are doing things a generic prompt can't.
• Your data gets cleaner. As you connect AI to actual workflows, you discover the gaps in your folder structure, your naming conventions, your templates. Each fix makes the next AI workflow easier to ship.

Both of those compound from the moment you start. Neither one happens for the firm that's still in the "we should look at this" phase 12 months from now.

The honest question

The 79% number isn't really telling you about AI. It's telling you about a competitive landscape where almost everybody agrees on the risk and almost nobody is moving.

That's the rare moment in any industry. The signal is unanimous, the action is missing. Whoever moves while the gap is open captures it.

Where are you right now — still talking, or already testing?

If it's still talking, the four steps above are the entire path. There is no other entry point. No vendor pitch will get you there faster than picking three workflows tomorrow morning.

The firms that win the next 24 months are the ones that ran the experiment this week.