Built agentic work-order dispatch, became company's head of AI
“A vision, unless it's backed up by action, it's just a hallucination.”
Cal lives in Toronto, originally from Yorkshire. He works as a service manager and project manager inside the maintenance division of a contracting business, with a prior life across tech startups and accelerator programs going back to 2016. He had been towing the line between the built world and the digital world for the best part of a decade, and the digital side kept calling him back.
The day-to-day was full of work orders. Every time a key account sent in a request, Cal was copying and pasting the description, title and job number into Google Sheets, deciding whether it was a warranty job, a project, a critical service call or a priority service call, then chasing down a subcontractor. The job itself was 15 minutes of admin per work order. The reality was that 15 minutes got smeared across two or three hours, and on a bad day a Friday night request might not get opened until Monday.
"If the customer says that we have an emergency and we need a plumber here within 4 hours, if it took me 4 hours to even open that email and figure out if they need someone, we've already let them down." In this industry, customer complaints almost always trace back to the same root cause: nobody picked up the phone in time.
He had dabbled in Airtable and Zapier going back to 2021, even written custom scripts to push them further. He was decent at it. He also remembered working until midnight on what was effectively the earliest version of vibe coding, thinking, "man, I wish that this could be a job." What he had not done was wire actual AI agents into any of it. Before joining, he sat down with the change management consultant on retainer at his company and pitched the idea of replacing what wasn't truly required of him. He got buy-in. That gave him the conviction to invest in the mentorship.
He almost talked himself out of joining anyway. "Is this just another shiny object? Am I being a sucker here?" The cynicism was real. Then he flipped it.
The one sure way to make sure that it doesn't work is to be cynical and not do anything, and just give the ball a kick.
The system Cal shipped is a full work order triage and dispatch pipeline. A Google Apps Script sits on his inbox and watches for work order emails. When one matches the criteria, it applies a label, which fires the n8n workflow. From there, an LLM agent reads the description and classifies the request — warranty, project, critical service call, priority service call — and creates the matching record in Airtable.
Photos were a fight in their own right. The industry runs on visual evidence, but the formats coming off site phones don't play nicely with most tools. "I needed a way to handle photos, which was a whole thing, if you remember, Tom." Dropbox didn't work. Google Drive permissions didn't work. Cal eventually routed images through Convert API to generate clean URLs that the rest of the workflow could actually use.
To keep his team comfortable while the system earned trust, he layered in a human-in-the-loop step. An agent posts into Google Chat with the work order, the description and its proposed classification, and waits for a thumbs up. Once approved, the workflow cascades through Airtable, filters the labourers and technicians by trade and availability, and surfaces the best matches. Cal picks one. The dispatch goes out by email and a Twilio text message lands on the subcontractor's phone.
That classification step was the moment the whole abstraction clicked. "Until that point, anything on the internet was apparently agentic. I didn't really understand what they were talking about. But now I do."
Mid-build problems were where the mentorship paid for itself the most. The photo handling problem alone could have stalled the project for weeks. Working through it together meant try, fail, try again, ship. Cal called this out by name as part of the value: "I believe in you, let's get through this together, let's try this. Okay, it didn't work, okay, let's try something else."
A category of work that could take up to 48 hours to even acknowledge now flows through in minutes. The dispatch bottleneck has moved up the funnel. Cal can now point at customer acquisition as the next thing to automate, rather than fighting the inbox. For a contracting business, where service complaints almost always trace back to nobody picking up the phone, that is a competitive lever the company can sell on.
The bigger shift is positional. Construction and maintenance companies do not have CTOs. They are starting to need a head of AI. Cal is presenting at the quarterly town hall this week as exactly that: the person interviewing every department, prototyping inside their workflows and freeing them up for the work that actually requires a human. The subcontractors he dispatches to — small plumbing companies, hood cleaners, the guys with a Debbie in the office — are already asking him whether he can build a version of the system for their own shops.
Probably for one of the first times in my life, I feel like I'm in control. I have the agency to make that happen.
He frames the investment as a vote towards the future he wants. The 15-minute work order was never the prize. The prize was being the operator who can stand in front of the owners with a working agentic system, and use that as the leverage to walk into the next set of rooms. "Whereas I was working on a 15-minute thing here, I'm now working on something that could unlock tens of millions in revenue on the next project."
A vision, unless it's backed up by action, it's just a hallucination.
Cal called out three things specifically: that I'm on every single call, that the calls don't get cut short when something juicy comes up, and that the group is small enough that you don't have to fight for five minutes of attention. He compared it directly to a startup incubator he and a buddy went through during COVID, where everyone was scrapping for their pitch slot and the rest was prerecorded. "That was not it. But this was not that."
The other thing he keeps coming back to is the 1% marginal gains mindset. Not the big breakthroughs. The boring daily wins: figured out the LLM, figured out the system prompt, figured out Convert API. He pushed momentum over perfection from week one, and that was the shape of what got shipped.
If Cal had not joined, in his words, he would still be on the diving board. "You're petrified to jump in and you're just running this over in your head, like, is this the thing? Is this what I should be doing? Meanwhile, everyone else has gone into the shallow end and they're progressively working their way to the deep end." The mentorship was the kick into the water.
Cut a 2-day document workflow to under 2 minutes
Read →Rebuilt a 7-year content workflow into a system that runs in minutes
Read →Launched solo AI agency, 7 demos and first client in 30 days
Read →