For AEC, Intelligence on Tap Is Your Defensible Moat

Your firm has deployed an AI chatbot, maybe even switched on the standard connectors in its general-purpose assistant. Your mail and cloud drives are a few clicks from being searchable.

You feel like you're making real progress, that true AI transformation is close because your people, who were not trained how to design prompts or understand workflows, are "using AI".

Then someone asks a real question. What did we learn on the last three jobs for this agency? The answer comes back fluent and cited, pulled from whichever systems happened to be connected.

Whether it is right depends on a decision nobody made. Of the several places that could hold the answer, which one does your firm trust, and who decided?

Stop at individual connections and your people get a faster search across everything, while the trust stays with the individual who may not remember why they decided.

Standard connectors are the easy part. The hard part is the definition: how a piece of work runs in your firm, who touches it, and where the answer lives at each step and in what system. A firm with that definition can make its own knowledge arrive on demand, inside the work. A firm without it has bought individual speed that never reaches the business.

What do standard connectors give your firm?

A connector is a socket. It gives an assistant a way into one system. It does not say what the work needs from that system, at which step, for which person, or which source wins when two of them disagree.

Someone has to decide that. In the assistants' own tools, that someone is you: you describe the job, and you supply the server for anything that is not a standard connector.

Two very large AEC firms built on records they already owned. Bechtel put its own equipment documentation behind a tailored model so people can query it instead of reading the manuals. Skanska's Operational Risk sidekick draws on thousands of case studies to suggest risk mitigation on jobs under way.

David Wilson, manager of functions at Bechtel, is quoted in ASCE's Civil Engineering Source saying the firm's library "has to be better tailored for the end user who needs it in real time and with information in the right context." He starts from the person who needs the answer and the moment they need it.

Both firms are very large, and this is reported practice rather than a measured result. What transfers is the approach: the raw material was already theirs, and each build was shaped around a job.

In the Engineering Management Institute and ActionsProve 2026 survey of more than 300 AEC professionals, 88 percent said they use AI tools at work and 21 percent said they use them daily, all self-reported.

The report sets that against industries where AI is embedded in daily workflows, where more than half of professionals use it every day. It also found that no respondent who used AI less than daily reported transformational productivity gains.

My read: a tool someone has to remember to open gets used when they remember, and a step built into the workflow gets used every time the work runs.

What does it take to make your firm's knowledge arrive on demand?

Three pieces, and each one stands on its own.

First, the definition. An AI Workstream takes one workflow and defines it end to end: how it runs today, who touches it, where the answer lives at each step, what systems hold critical data and which steps a model can take, which need a person and which should not exist.

It ends in a scoped recommendation for what to build first. The definition is yours to build from, on any team or any platform.

Until someone does that work, five people describe five different processes, all of them true and none of them complete. After it, the firm has a baseline, a ranked list of where the time goes, and an understanding of the AI impact on that workflow.

Second, the connections, built to that definition. Because a workflow definition says which system holds the answer at each step, each connection pulls from the ones your firm trusts, not from everything that happens to be reachable. We build those connections to the systems the workflow crosses, including the ones with no standard connector.

Third, a record that stays yours. UniversalContext™ is the organizational intelligence platform that holds it, and more importantly, protects it. The platform ties people to projects and contracts to clients, resolves duplicates, flags conflicting information across documents, keeps past answers so the next question starts further along, and is not tied to one model vendor.

AI models get smarter for everyone. Your organizational intelligence gets smarter only for you, and that advantage compounds every day.

A competitor can switch on the same connectors. They cannot switch on your definition. That is the moat.

Why has this not happened in your firm? The standard connectors were the part an administrator could switch on, so that is where firms start. Nobody owns the definition, and until someone does, the assistant answers from whatever it can reach.

The principal who knows how a proposal gets approved, the project manager who knows which subconsultant needs a phone call, the controller who knows which report to distrust: each holds one piece, and nobody has asked any of them to write it down for someone else.

Felsburg Holt and Ullevig (FHU) wrote a five-year strategic plan before generative AI existed. What it bought was not foresight. It was knowing what the firm was and what it did, and that clarity is what let FHU act when AI arrived.

That was a strategic plan, not a workflow definition. Five years is how long that engineering firm has been at this. It is not how long a firm needs to start.

Start with one workflow, the one everyone complains about and nobody can fully explain. Ask the people who do the work, not only the people who own it. For each step, write down two things: where the answer lives, and who decides when two sources disagree. That list is the first draft of a definition, and you can write it this week without buying anything.

If you cannot finish it, that is the finding, and it is what our AI Workstream engagement is for: three to five weeks, one complete workflow, the fee is fixed, and the output gives you a clear understanding of how to begin building your unique organizational intelligence.

Here is the test I would run first: name the workflow you would hand a new principal on day one. Could you tell them, step by step, where the answer lives and who decides? If you could not, the AI assistant will not know either, and every connector will carry that gap into your answers faster.

That answer gets harder to write down every time someone who holds it leaves, and no connector brings it back. Start with that workflow this month.

---

Where to go next: an AI Workstream is one of the four bounded engagements. It takes three to five weeks, defines one workflow end to end, and ends with a scoped recommendation for what to build.

Holly Buck, President of FHU, and Chad Sanderson of Universal Mind take this up on stage at the ACEC Fall Conference in Phoenix on Monday, October 26, in "Ignite Innovation: Moving from AI Skepticism to Organizational Intelligence." FHU is a reference customer of Universal Mind and UniversalContext.