94% vs 27%: Why AEC Is the Outlier in the AI Adoption Story
94% of large enterprises have touched AI in some form. 27% of AEC firms actually use it in workflows.
That gap is the real story of 2026 in construction.
Where the numbers come from
The 94% figure comes from the Stanford AI Index 2025 — one of the most comprehensive annual surveys of AI adoption across industries and geographies. At that penetration rate, basically every Fortune 500 has a pilot, a vendor relationship, an internal task force, or a chatbot deployment somewhere in the org.
The 27% number comes from ASCE's adoption data on the AEC sector specifically ([SOURCE_URL]). And even that figure is generous, because most of those 27% report using AI for one narrow task — often a document chatbot, occasionally an image-based defect detection tool on a large infrastructure project. Embedded AI across multiple firm workflows, with measurable output improvement, is a much smaller number.
Two industries. Completely different curves.
What the curves actually look like on the ground
What I've watched playing out at AEC firms over the last 18 months isn't a single story. It's three stories running in parallel.
The firms that adopted AI early have it embedded in proposal workflows, contract review, and project documentation. These are typically the firms where one director or senior engineer took AI seriously before it became an industry conversation, built something internal, and then systemised it firm-wide. They didn't wait for a vendor to package it. They built the scaffold around a general-purpose model and made it specific to their work. Now they have tools that know their contract families, their client base, their typical risk exposures. That institutional corpus is compounding month over month.
The firms in the middle have one or two engineers experimenting on the side, with no firm-wide rollout. This is probably the largest group. Capable people inside the firm have discovered that AI makes parts of their job faster. They use ChatGPT to draft specification sections. They use Claude to summarise long reports. But none of it is systematised, none of it is shared, and none of it is building toward a firm-level capability. The firm isn't getting dumber, but it's not getting smarter either.
The firms at the back haven't started, and most of the leadership team treats it as a 2027 problem. The reasoning is usually: "we'll wait until the tools mature," or "we need to see regulation first," or "our clients aren't asking for it yet." All of these are defensible positions in isolation. Together they amount to strategic patience that's compounding into a capability gap.
Why this isn't like the BIM transition
The thing the back-of-the-curve firms keep missing: this isn't like the BIM transition where you had ten years to catch up.
BIM adoption was slow because it required capital investment in software licenses, hardware, and training. It required client-mandated adoption on specific projects before it became a general expectation. The moat for BIM adopters was limited — the tools were packaged, the skills were trainable, and a firm that started late could largely replicate what an early mover had built.
AI workflow advantages compound month over month, because the firms that adopted early are now training their tools on more of their own work. The moat isn't access to the technology — every firm can buy a Claude or ChatGPT subscription tomorrow. The moat is the corpus you've built up around it: organised project archives, clear naming conventions, a senior who's already iterated through the bad prompts.
A 15-person firm starting today is competing with a 15-person firm that started a year ago and now has a tool that knows their actual project history. That's a real gap. It widens every quarter the back-of-the-curve firm waits.
The honest framing for directors still on the fence
You don't have to pick the right model or the right vendor. You have to pick one workflow and start.
The 27% who report using AI didn't get there through a rollout strategy. They got there because someone inside the firm refused to wait — picked one painful, time-consuming task, asked whether AI could help, and started iterating. That's the entire playbook. One workflow. One person who cares enough to iterate. That's what separates the 27% from the 73%.
The firms still in the 73% that don't are mostly waiting for permission. Permission isn't coming. The industry won't standardise on a single AI vendor the way it standardised on Revit. The regulatory clarity firms are waiting for applies to high-risk AI decisions, not to the document drafting and knowledge retrieval tasks that are the actual entry point for 90% of AEC firms.
The cost of starting is a subscription and a few weeks of a senior's time. The cost of waiting another year is a growing gap against the 27% whose tools are getting smarter every month.
Where does your firm actually sit on that 27% vs 94% line — closer than you think, or further?
I write about AI for engineering and construction firms weekly: → Full breakdown: https://sigmametrix.net/insights/stanford-ai-index-27-vs-94-aec → Free AI Readiness Audit (7 questions → your 2-page playbook): https://sigmametrix.net/audit → Newsletter for AEC firm directors: https://sigmametrix.kit.com/8686be4583