by /
I met Rick DeMedeiros in Atlanta, and within five minutes he’d handed me the clearest example of AI agents for service businesses I’ve seen all year. The first thing I told the camera was that he doesn’t look like a lawyer. The photo on his website is the traditional guy in the suit. The real Rick — the one sitting across from me at The Westin, eating steak with some green sauce neither of us could name — is warm, plainspoken, and a little restless. He runs a small workers’ comp and personal injury firm up in Alpharetta. The brand is “Pick Rick.” He is, by any definition the marketing industry uses, a boring local business.
That is exactly why he is the most interesting AI story I’ve told all year.
What AI agents for service businesses actually do
We had just come off the stage at DigiMarCon, where I ran a live AI agent on his business in front of a room. And the moment Rick keeps coming back to isn’t the flashy output. It’s what he noticed sitting next to me while I worked. “The number one aha moment for me,” he said, “was how specific you were, how much of a plan you had.” He watched me talk to the agent — not type at it, talk to it, ums and ahs and all — the way you’d brief a smart new coworker.
“This is everything we know about Rick. Here’s our goal. Here’s how we like to do the work.” Not micromanaging it line by line. Giving it the big goal and the context, then letting it run.
And it ran. For about forty-five minutes it worked through a plan we could watch unfold: research everything about Rick, look at the $20 million case, study how his years on the insurance-defense side make him sharper on the workers’ comp side, map his competitors, find the trust marks he isn’t leveraging, figure out what content the website is missing, tune the ads. I like the token math on this. When an agent comes back in two minutes, it didn’t work hard. When it grinds for forty-five and burns a few dollars of tokens, that’s the labor fee, and you can feel the difference in the output.
What came back wasn’t a wall of text you have to dig through. It was formatted, visual, with diagrams and a plan — green is good, red is bad, here’s what to fix. And crucially, it was already action. That agent had updated our Basecamp, created the tasks, and handed the next steps to real people on the team. I keep saying it and I’ll say it here: our documents are better than most AI output because our documents are designed for action, and designed for human eyes.
The magic box only pays out what you feed it
Here’s the part I want you to sit with, because it’s the whole point. Rick looked at all of that and said it looked easy. And I said back to him: that’s exactly what your clients say about you. When someone who’s a pro handles the insurance company, navigates every hoop, and gets the result, it looks easy — because they’ve done it a thousand times and failed and tried again and failed and eventually got good. The agent looked easy for the same reason the master looks easy. The skill is invisible. The specificity is invisible. What’s actually happening underneath is years of judgment being poured into clear instructions.
I have a mental model I gave Rick at the table, and I’ll give it to you. Imagine a box, and whatever you put in it, you get ten times more out. Put in a pen, ten pens come out. Put in a hundred-dollar bill, you get ten of them. That’s what technology is. The magic box just got fancier, but it’s the same magic box — it’s a multiplier.
So the only real question is what you put in. Rick has decades of real cases, real stories, genuine care for injured people who can’t afford to get squashed by an insurance company. That’s rich input. Multiply that and you get something powerful. But if an eighteen-year-old with no direction feeds the box “I like to play video games,” the box gives them more video games.
The unfortunate thing is that people blame the box. I blame the input. That is the whole game with AI agents for service businesses: the output is only ever as good as the judgment you feed in.
Which is also why I think the whole “prompt” conversation is broken. When ChatGPT blew up, the internet-marketer crowd started selling prompt packs — “100 prompts for $99, now you’re an AI expert.” It’s bogus. The word prompt already misdefines the problem, because it assumes there’s some magic phrase, some abracadabra that unlocks the gold chest. There isn’t. It’s your ability to think, reason, and communicate.
Rick got this instantly, because it mirrors his own work. When he sits with an injured client, is he “prompting” them? No — the second you think of it that way, you’ve already lost. What he’s actually doing is bringing empathy backed by understanding and experience, injecting motive and context into every direction he gives. That’s what any service business worth its weight already does with people. It turns out it’s exactly what works with agents too.
Care is the moat no model can copy
So the real trick isn’t a phrase. It’s that you have to treat AI like a human. I say please and thank you. I use full sentences. I onboard it like a new employee — give it access, let it try a task, give it feedback the first time it gets something wrong, and watch it get better and remember how we like to do things.
The reason people who actually get results from AI all say the same thing — “I treat it like a person” — is that the humanity component is the skill. Your ability to manage humans is your ability to manage agents. If you can’t give a worker clear directions, you can’t blame the worker. Same with the machine.
And here’s the moat, the thing no model can copy. Rick’s edge isn’t legal trivia — you’d never have enough knowledge to quiz him on Georgia workers’ comp law anyway. His edge is that he genuinely cares, and in a field where the bar for lawyers is famously low, care is a competitive advantage. He told me clients come in hurting physically and financially and leave in a better place, and that’s the whole job.
People can smell whether you mean it. The AI backlash is real, and it cuts in his favor: the more convincing the deepfakes get, the more people crave a human who actually gives a damn. What technology does is make his care legible to the world — through Google reviews, through a track record, through the fact that ChatGPT and Claude and every other engine are now basically judges, scanning for enough real signal to tell an injured Spanish-speaking worker that this small firm is the one to trust. The agent didn’t replace Rick. It multiplied him. That is exactly what AI agents for service businesses are for.
That’s why I’ll take a “boring” injury lawyer over a shiny tech demo every time. The lesson generalizes to every roofer, every dentist, every local shop that thinks AI is for someone else.
If any of this landed, don’t buy a prompt pack. Learn by doing the real work. That’s the entire premise of what we do at BlitzMetrics — we put nineteen-to-twenty-two-year-olds on actual paid client accounts, businesses exactly like Rick’s, and let them learn to onboard an agent the way you’d onboard a teammate. Done right, AI agents for service businesses don’t replace anyone — they multiply the people who already care. Watch the full episode, then go put something worth multiplying into the box.