Our AI Agent Clocked In This Morning. We Billed the Client for It.
#CrackedEngineering Ep. 1 with Kenny Shultz
You’re reading People, Process, Technology. We write about what’s actually working inside AEC firms right now. Not theory. Not vendor pitches. Just what we’re building, testing, and learning. If someone forwarded this to you, subscribe here.
In software, they call them cracked engineers. The 10x developers who take extreme ownership, stay optimistic, and apply themselves to the craft with something close to religious fervor. For decades, that concept didn’t translate to the built environment. You can’t 10x a concrete pour. You can’t ship a building in a sprint. The physical world has constraints that software doesn’t, and those constraints kept a ceiling on how much one engineer could multiply their output. Until now.
Artificial intelligence is removing that ceiling. For the first time, engineers who design, inspect, and build the physical world can operate at a level that used to be reserved for people writing code in Silicon Valley. #CrackedEngineering is a show about that shift. Kenny Shultz and I get on the mic every couple weeks and riff about what’s actually working in AEC right now. Not theory. Not what some vendor promised at a booth. What’s working today, in real firms, on real projects. This is the first one.
Kenny Shultz is the Engineering Director of PermitZip, and co-host of the Blueprint Tour. He’s also one of the most technically aggressive AE firm leaders I know. What I mean by that: while some firm owners are still debating whether to let their teams use ChatGPT, Kenny is billing clients for AI agent time.
And he’s put it in his contracts. Not debating about it theoretically at a roundtable. Actually doing the thing.
People
Here’s what happened. About three weeks before we recorded this conversation, Kenny’s team wired up Claude with MCP tools, skills, and a custom agent connected to their own internal systems. Dropbox. Slack. Transcription. Project files. Time tracking. Everything.
The adoption was fast, but it wasn’t a light switch. Kenny’s team has a range of experience levels, from junior engineers to senior technical staff with decades of traditional workflows behind them. Some people got it right away. Others needed coaching and mentoring, just like they would with any new tool or process. Kenny had to teach his team what skills are, how MCP tools work, what an agentic workflow even looks like. And he was figuring it out in real time himself. You can’t hand someone a manual for something you’re still building.
That’s the part that gets lost in the AI conversation. The technology moved fast. The people part still took leadership, patience, and a willingness to walk alongside your team while they got comfortable. Kenny said something that stuck with me: you almost have to teach people the new workflow of engineering as you develop the new workflow for engineering. Those two things happen at the same time, not sequentially. Flying the plane while building it.
What surprised him was how quickly the culture shifted once a critical mass formed. Even the senior engineers, the ones you’d expect to push back, went all in. They were commenting on it. Engaging with it. Using it daily.
Kenny told me that when he walks through screen shares with his engineers now, the first screen they go back to is Claude. Not Revit. Not a Bluebeam PDF. Claude.
That didn’t happen overnight. Kenny has been pushing this direction since early 2023. He told his software team to start using AI to write code. The director at the time said no, not really. One developer quit because he was deliberately not using it, even after being told it was a job requirement. His reasoning was fair. He was worried about skill atrophy, about future employers who might prohibit it. Kenny understood. But he also told him: you either need to quit or get fired. That sounds harsh. But Kenny saw where this was going before most people did.
Three years later, his entire team is running on it. Why?
It’s a mindset and culture thing. I have mentors ranging from their mid-fifties to late seventies who are sharp and into this stuff. And I know 22-year-old engineers who say it can’t draw a line in CAD better than they can. Right people, right seats.
Because they see their leader is hands-on. Not just talking about it or dictating that “we need to be using AI more”. Kenny is in the trenches with them.
The technology isn’t the bottleneck. People and culture are.
Process
So here’s the part that got my attention. Kenny’s firm is now tracking AI agent costs at the project level and converting them to billable hours.
The contract language reads like this:
“All billable time under this agreement, including time attributable to AI tools, is measured, tracked, and invoiced in hours of professional service. Agent runtime, compute time, and AI-assisted task completion are converted to equivalent professional service hours using the firm’s internal methodology.”
That’s real contract language. In a real engineering firm. Today.
Here’s how the math works. When Kenny’s agent runs a task, it costs real money in tokens. A quick project search might run 60 cents. A deeper analysis might hit a couple bucks. Kenny applies a 10x multiplier to the token cost. So if it costs 27 cents, the client sees $2.70 converted to an equivalent hourly rate.
His logic: the 10x isn’t a money grab. Those tokens are worth more because a licensed professional engineer is driving the agent. The expertise behind the prompt is what creates the value. Not the compute.
There’s an old story about a ship repairman who fixes a broken engine with a single tap of a hammer and bills $10,000. The itemized invoice reads: $2, tapping. $9,998, knowing where to tap.
Same idea.
Kenny told me about a recent project where the architect put massive skylights everywhere. As a mechanical engineer, he saw it immediately. Red flag. So his agent ran a thermal calculation, found product specs online, wrote a Python script to model the heat gain coefficients across different glass options, and determined the skylights would double the tonnage on the mechanical system. Then it saved the report, noted the study on the timesheet, wrote the email summary, and flagged it for the schematic meeting.
That used to take four hours to even begin. The agent did it while Kenny moved on to something else.
And here’s the thing. Even before Kenny bills for the agent time separately, the agent’s work log gives him something he never had before: a detailed, automatic record of everything that was done on the project. Every file reviewed. Every calculation run. Every email parsed. That level of transparency in an invoice is unheard of in consulting engineering.
Technology
So what made all of this possible? And why now?
Rewind to November 2022. ChatGPT drops and the world loses its mind. Kenny and I were both in it immediately. He started using it to write business reports and internal communications. I was doing the same on the structural engineering side. We weren’t building agents or writing skills. We were just asking it questions and seeing what came back. But we were building the muscle.
Through 2023 and most of 2024, it didn’t look like much had changed. The models got smarter, the context windows got bigger, but for engineers in the built environment, the day-to-day workflow was mostly the same. You could ask it questions. You could draft emails. You could get a decent first pass at a report. Useful, but not transformative.
Then things started stacking. Anthropic released Claude Code, which let developers hand off real coding tasks to AI from the command line. Then came CoWork, which gave non-developers a way to automate file and task management from their desktop. Then skills, connectors, and the ability to wire AI into the tools you already use. Each release built on the last. And Anthropic has been on a tear, shipping capability after capability at a pace that’s hard to keep up with even if you’re paying attention.
Here’s the point. None of that mattered if you weren’t already in the arena building the muscle. Kenny didn’t go from zero to billing agent time in three weeks. He went from three years of preparation to three weeks of explosive output. The recent convergence of MCP servers, skills, and custom agent frameworks was the spark. But the fuel had been accumulating since November 2022. That’s what we call aggressive patience.
Think about an exponential curve. The early part looks flat. You’re grinding, learning, experimenting, and it feels like nothing is happening. That’s 2023 and most of 2024. Then the curve bends. That’s where Kenny is right now. His team’s custom platform lets engineers talk to the agent by voice, search project context across dozens of conversations, run calculations, generate reports, draft emails, and log time. The agent clocks itself in when it starts working on a project.
The telemetry layer is what ties it together. Kenny can now see per-user, per-project AI spend in real time. On the day his team really started using it, one person hit $45 in a single day. The average is closer to $25 or $30. That’s roughly $600 per month per employee in AI costs.
Most firm owners would see that number and say no thanks. Kenny saw a revenue opportunity.
If each employee generates $600 a month in AI costs that can be converted to billable work at a 10x multiplier, that’s $6,000 a month in additional revenue per employee. For doing the same work, faster, with better documentation.
And the returns compound. Every meeting summary the agent writes gets stored. Every project search result makes the next search smarter. Every skill built for one project applies to the next ten. Kenny said it well: ideas used to die on the riff room floor. Now they get built in the same week they come up.
Kenny didn’t wake up one morning and decide to build an agentic engineering firm. He started building the muscle in 2022 when most of the industry was still figuring out what a prompt was. He experimented with every new release. He pushed his team even when they pushed back. He watched the ecosystem evolve. And when the pieces finally came together, his team was ready to move.
That’s the part most people miss. The compound effect only works if you start early enough for the interest to accumulate. The best time to start was 2022. The second best time is today. It’s not too late. But the curve is bending, and the gap between those who started and those who haven’t is getting wider every month day.
If your firm is starting to ask the same questions Kenny already answered, we’d like to hear from you. Just hit reply.
— Nick and Kenny