Walk into any architecture or engineering firm right now, and you will hear the same debate. Can AI actually understand a building? Who is liable when it gets something wrong? Is any of this genuinely useful, or just hype in a chatbot interface?
Fair questions. They are also questions our industry has asked before… twice. Both times, the skeptics ended up on the wrong side of history.
The Ghost of CAD Past
Forty years ago, the industry was transitioning from hand drafting to CAD, and the pushback was fierce. Veterans argued that computers could not replace the craft. And honestly, they had a point at first.
Early CAD was painfully slow and expensive. Using AutoCAD in the late eighties, it was possible to exploit the REGEN command to create animations; you could draw something, hit regen, and because the refresh was so incredibly slow, the redraw itself became the animation. That is how sluggish the tools were in the beginning.
Then hardware got faster, the software got smarter, and CAD decisively won.
Then Came BIM
Twenty years later, we ran the same play with BIM. When Revit first launched, it was rough: crashes, bugs, and architects struggling to get drawings out the door. In many cases, it was not meaningfully better than CAD, just newer and more frustrating.
Here is where it gets interesting: the same people who had championed CAD over hand drafting were often the ones now resisting BIM. The disruptors had become the defenders. They were not wrong about Revit being buggy; they were wrong about what that meant.
Again, hardware caught up, the software matured, and BIM is now clearly winning in the industry.
Now It Is AI’s Turn
Three years ago, it was easy to dunk on AI. The hallucinations were comical, the outputs clearly unreliable.
Here is a concrete example from something we discovered internally. In May 2025, Claude Sonnet 4.0 was tasked to review electrical panel schedules following the rule: spare breaker capacity shouldn’t fall below 10%. It was fed into a panel with 14% spare breakers, and the Claude model confidently declared it was below threshold. This was obviously wrong, and embarrassingly so.
Less than a year later, in February 2026, Sonnet 4.6 came out. Same review, correct answer.
That is not an anecdote about one model getting better. That is the curve; the same curve we have seen twice before, and it is closing the gap between “confidently wrong” and “genuinely useful” faster than most people in our industry realize.
Skepticism Is Not Dismissal
Be skeptical. Challenge these tools and make them earn a place in your workflow. Architects and engineers design buildings that people live and work in, so “the AI said so” is not an acceptable standard of care.
But do not confuse skepticism with dismissal. Those are very different positions, and our industry has a habit of blending them at exactly the wrong moments.
Forty years ago, the people pushing CAD were early… and right. Twenty years ago, the people pushing BIM were early… and again, right. Every time AEC goes through one of these transitions, the early criticisms are legitimate: the software is immature, and it does not work as well as established tools. And every time, it is proven that the new technology wins out anyway.
It Is Okay to Not Adopt AI… Yet
You don’t have to go all-in on AI. You genuinely cannot, because there is not a coherent “it” to go all-in on. Anyone telling you they have figured out the complete AI strategy for an architecture firm is selling something.
But ignoring AI is a different bet entirely. That is a bet that this time is different, that this particular technology curve will flatten where the last two did not. History suggests that it is not a bet worth making.
Join Our Free Webinar
May 19, 2026 • 1:00 ET / 10:00 PT
“Create Single-Line Diagrams in Revit Without Late Nights or Change Orders”