Your People Got Faster. Your Firm Did Not.

Ask a room of engineering firm principals how many of their people use AI and the hands go up fast. Nearly everyone.

Then ask the second question. What does your firm know today that it did not know a year ago?

I have watched principals work that one out loud. There is a pause, then a list of things individual people can now do faster, then a slower recognition that the list is not an answer to the question.

Somewhere in the firm, someone has already worked out the thing the next project is about to run into. Nobody can find them, and the firm will pay to work it out twice.

Those are two different measurements, and this profession keeps reporting the first as though it settled the second.

The distance between an individual getting faster and an organization getting smarter is the whole game, and almost nobody in this industry is measuring it, which is a different problem from the gap between using AI and adopting it.

What does an adoption rate actually measure?

Start with the sharpest numbers in this profession. The Engineering Management Institute and ActionsProve surveyed more than 300 engineers and architects for their Future of Work Report in Engineering and Architecture 2026. Of those, 88% use AI tools at work and 21% use them daily.

Even among daily users, transformational gains are a minority experience. Among less than daily users, none report transformational gains.

Not a low number. Zero.

Three more industry reports seem to contradict these numbers, and each other, on adoption. They do not. They asked different questions, and every one of them still misses the thing that matters.

Who asked	What they actually asked	The number
Deltek	Have you adopted AI, and is the impact measurable?	70% adopted, 38% see measurable impact
Unanet	Do you use AI, and do you trust the data behind it?	75% use it, 29% trust the data
Bluebeam	Do you use AI for automation, problem-solving or decision-making?	27%
EMI and ActionsProve	Do you use AI at work, and how often?	88% at work, 21% daily

Publisher, date and sample for every figure above: sources.

Bluebeam's 27% looks wrong and is not. It asked the narrowest question on the list, of a different room: technology decision-makers across five countries, not Unanet's AEC leaders or Deltek's A and E firms. Its fieldwork is the oldest, July 2025.

Deltek and Unanet sell software into this market, so they know it. Their numbers are honest and targeted. None of these reports measures what the firm is actually gaining.

McKinsey's state of AI in 2026 puts it in three numbers rarely printed together. Respondents whose own productivity improved: 80%. Respondents attributing any EBIT impact to AI: 37%. Respondents clearing McKinsey's high-performer bar, at least 5% of EBIT attributed and significant value reported: 6%.

Those are not three cuts of one number. The first measures individuals on their own work; the other two measure organizations' EBIT. The same study finds an association it does not explain: nearly three-quarters of high performers have redesigned workflows, against one-quarter of everyone else.

Which means a firm of individually excellent AI users is still a firm where every person works from their own private picture of what is true.

This is not an argument against training people. Skill is real and it compounds, which is why the daily users see returns. More skill makes each person better at the tools. It does not give them a shared picture, and the difference between those two things is what an adoption rate cannot see.

What does the organization actually keep?

Here is a pattern we keep seeing across engagements. When a firm says it is AI-forward, it usually means it has deployed tools: chat seats, a copilot, a vertical product or two.

It rarely means that knowledge is connected, or that anyone could say what the return has been. AI-forward usually means AI-scattered.

The question we ask: if a project manager works something out about an agency's review process today, how does the pursuit team find out tomorrow? The honest answer is usually a hallway, and a hallway does not scale.

What a tool learns stays inside that tool. What a person learns leaves with them. ACEC's Research Institute found that in 2022 roughly 184,000 engineers retired or left the United States profession while 166,000 new graduates entered. One year's snapshot, not a standing annual rate.

The usual read is that firms must do more with fewer people. There is a better one.

Nobody went to engineering school to spend a Tuesday hunting for a decision somebody made on a similar corridor four years ago. A thinning bench makes that hunting worse, because the person who knew is the person who left.

Give your engineers back the hours they lose chasing what the firm already knows and you do not get a smaller team. You get the team you hired, doing the work they trained for.

This is why the adoption number keeps climbing while the firm-level picture holds still. You have been measuring the wrong thing, and the thing you have been measuring is genuinely improving, which is what makes the illusion so hard to escape.

Individual productivity is real. It shows up in how a proposal gets drafted, how a memo gets summarized, how fast someone reaches a first pass. It does not show up on the income statement, because none of it accumulates anywhere the organization can reach.

Skill compounds for the person who has it. What the firm knows compounds only if the firm can reach it.

An engineering firm is, in the end, a machine for turning what it has learned into what it can charge for. Every project teaches it something: what the agency actually wanted, why the schedule slipped, which assumption did not survive contact with the field.

Right now most of that is being learned by individuals, at speed, with better tools than they have ever had, and then it walks out of the building at five o'clock. The firm is not getting smarter. It is getting faster at being exactly as smart as it already was.

Stop reporting adoption to your board. Report the lines that only move when the firm itself gets smarter: net revenue per employee, pursuit throughput and win rate on the work you chose to chase, and time to resolve a regulatory change across every active project.

That is organizational impact. That is value extraction. An adoption rate doesn't translate to value, and your employees speeding themselves up with AI does not provide organizational intelligence.

Ask the hallway question out loud at your next leadership meeting. If a project manager works something out today, how does anyone else find out tomorrow? Watch how long the silence runs. The length of that silence is the measurement.

Then pick one repeating program: a corridor study, a statewide on-call, a client you serve every year. Ask what the firm learned on the last three. Not what is in the files. What was learned.

If the answer requires interrupting two specific people, you have found the thing worth fixing, and you found it without buying anything.

The firms ahead in three years will not be the ones whose people adopted AI first. Nearly everyone has. They will be the ones who built organizational intelligence, not just individual productivity.

In a business that sells judgment, that is the whole difference.

Holly Buck and I take this up at the ACEC Fall Conference in Phoenix, Monday, October 26. Come argue with us.

What does your firm know this morning that it did not last October? If it takes more than a minute, that is the answer.