Sign in Subscribe
Leadership Edge

What AI Can Actually Do in a Real Firm Today

📰 Eleven AEC vendors, one question: what can your AI actually do in a real firm today? The honest answers were more useful than any demo.

What AI Can Actually Do in a Real Firm Today

If every vendor in your stack is telling you their tool is agentic now, how do you tell which ones actually are? And what does that even mean?

💡
Who this episode is for: Especially relevant if you're the one fielding AI pitches from every vendor in your stack (and beyond) and need a way to separate what's real from what's marketing.

Summary

For this episode of The Next AEC Tech Stack, From Autocomplete to Agent: Where AEC Actually Is (TRXL 241), I asked every partner in Arc 2 the same question: in a real firm today, what can this actually do, and what can't it? Not the roadmap. Not the demo. What's actually shipping and actually working on a live project right now*.

*Disclaimer: Keep in mind that everything they said could change tomorrow (in fact, it might have already changed since many of these interviews were recorded weeks ago!). Things are moving fast.

I talked to representatives from Chaos, UpCodes, Twinmaster, Egnyte, Bentley, Graphisoft, Skema, BIMbeats, Finch, Autodesk, and AVAIL, plus three practitioners who have to live with what these companies ship. The answers ranged from "we don't really train it, it's discrete math" to "the harness is 90 percent of the work," and a few people were candid about where their tools are today that may not be meeting expectations.

What came out of it wasn't a tidy verdict. Rather, it was a pattern: the firms and vendors getting real traction aren't the ones chasing the most impressive demo. They're the ones who've figured out exactly where the deterministic, rule-based work ends and the genuinely hard reasoning problem begins, and who are honest about which side of that line they're actually standing on.

This series brought to you by
Graitec

Graitec provides AEC software, training, and expert support to help architects, engineers, and construction teams design, build, and manage projects more effectively.

With support from
AVAIL

AVAIL's content management system (CMS) helps architecture and engineering teams find, manage, and reuse their Revit, AutoCAD, Civil 3D, and other technical content in one place.

💡
Free and paid subscribers are getting the same full newsletter for this series. Normally the full edition of Leadership Edge is a paying-member benefit. For The Next AEC Tech Stack it isn't: every edition goes out complete to everyone on the list, free accounts included, along with the Field Report that goes with each one. The partners above paid for the series so the industry could actually use it.

If someone forwarded this to you, sign up free and the rest of the series lands in your inbox as it publishes. TRXL+ members also get the ad-free feed, the full back catalog, and all the content that goes back behind the paywall when we return to our regular schedule following this series.
Catch the full episode

What's Inside

  • The wrapper problem. Many tools calling themselves AI are just code wrapped around an existing LLM, so what is the real, unexplored opportunity?
  • The 90/10 harness split. One guest put a number on how much of a real AI solution is the model versus everything built around it, and Autodesk's VP of Software Architecture had a pointed reaction to that number.
  • The token bill nobody saw coming. What happens when a firm deploys an AI tool across the whole company and the monthly bill hits six figures, and what that does to how firms will have to think about pricing their own work.
  • The alarm bell for a bad AI pivot. A practitioner explains exactly what tips him off that a product rushed into "AI mode" without solving its first problem.
  • What 2030 actually looks like. Several guests describe a firm that isn't smaller so much as reorganized, and one goes as far as predicting agent-to-agent procurement.

Read next