AGC Tech 2026: The Year AI Agents Started Getting Names

The Bootcamp: AI Stopped Being a Keynote Topic and Became Homework

Every year there's a session or two that tells you where an industry's head actually is, underneath whatever's trending in the headlines. This year at AGC Tech, that moment came from a talk about a piece of software named Wendy.

More on her in a second. First, the bigger picture.

Tuesday afternoon opened with the AI Bootcamp: From Experimentation to Execution, led by Dr. Armin Yeganeh and Zach Giglio, and it wasn't a lecture. It was hands-on: estimating, project management, field communication, documentation, the actual day-to-day functions that determine whether a construction firm runs well or doesn't, worked through directly with AI tools in the room.

That's a real shift from a few years ago, when "AI in construction" sessions were mostly someone on a stage explaining why you should care. Nobody at this bootcamp needed convincing of why they needed AI. It was all about sharing, giving tips and suggestions on improving the way you use it now.

Wendy

Wednesday morning's standout was Building the AI Back Office for Construction, from Lars Herman and Zach Koerber. Their thesis: 2026 is the year of the AI agent, and construction still doesn't fully trust it. The trick to building up a system that you can rely on is to structure the process first. Get the team to actually adopt it. Build trust with a human still in the loop. Then hand the work to an agent: one that knows the company, its standards, and every document that came before it, and doesn't forget any of it.

And at some point in the talk, they mentioned they'd named their AI agent Wendy. After Lars's mom.

It's a small detail. But it's also the whole point of the session in one sentence. The company is trying to bring in personalization and make the agent feel like part of the team.

The Reality Check

Not every AI conversation in Minneapolis was rosy, and that's a good thing. The AI Reality Check: Lessons from the Front Lines, a panel with IT leaders from two general contractors, was built specifically to cut past the marketing noise: what's actually moving the needle on forecasting, scheduling, and risk detection right now, versus what's still just a slide in someone's deck. They didn't dodge the harder questions either: data security, model accuracy, vendor verification, the governance side of rolling this out without creating a mess nobody signed up for.

That's the same instinct behind our own AI glossary eBook, honestly: the point was never to hype AI, it's to help people tell the difference between an agent that does real work and a chatbot with a paint job.

It's Not You, It's Me

One more session worth flagging: a panel on construction's interoperability problem, bluntly titled "It's not you... it's me." Designers, GCs, and trades still mostly run on disconnected systems, and the panel worked through real friction points of data that doesn't follow a project from handover to close-out.

A Few Other Things Worth Knowing About

"Kill Your PDF" (Jeff Sample) made the case that going paperless was never the finish line, treating a PDF as the end of a digital workflow, instead of just a format, is where a lot of "digital transformation" quietly stalls out.

AGC's Got Talent, a new format this year, had contractors pitching real solutions to real problems in five minutes flat, including a genuinely good one from Micah Davis at Broaddus Construction on turning individual employees' scattered AI wins into something the whole company can actually use, instead of everyone quietly reinventing the same prompt.

Building a Technology Operating System to Scale with Backlog (Shannon Tevendale, Faulconer Construction) made a point worth repeating: modernization doesn't start with picking software, it starts with a people-first look at how the work actually happens. Platforms come second.

The Thread Running Through All of It

Strip away the individual sessions and one thing was consistent all three days: this industry has stopped asking whether to use AI, and started asking how. Everyone wants to know how to build trust in it, how to avoid the hype, and how to get a team to actually adopt something new without a training program nobody has time for.

That's the exact bet we've made with ServiceIQ. Not AI bolted onto a feature list, but a real agent that runs the job, the same way Lars and Zach's back-office AI does: no training required, trusted enough to just get on with the work.

If the "LLM vs. agent" distinction that kept coming up all week is still a little fuzzy, we put together a plain-English breakdown of it below, worth a look. Or see it for yourself.