How Businesses Use Multiple AI Agents to Build Software
16 min · October 20, 2026
A commercial real estate team shows how a four-agent AI system cut a three-week site search to hours, and the group debates cost, risk and human oversight.
Key Takeaways
- AI automated a site search that once took three people three weeks, finishing it in about three and a half hours.
- AI can auto-complete tedious documents like letters of intent and contracts, which can then be scanned and corrected.
- Use AI as a lens to identify opportunities, not a crutch: you still need real sources for what you tell clients.
- A four-agent setup can turn half-formed ideas into precise prompts for a coding agent, working like a translator for a manager and a coder.
- AI tools like Claude Code can cost thousands of dollars a week, so check whether the savings are real.
- Keep a human at the helm and do not give AI sensitive information like financial data or access it does not need.
- AI tends to tell you what you want to hear, so rephrase questions and have someone else test results.
Main Topics
How several AI agents hand off ideas, prompts and code to build websites quickly.
Site searches, demographic reports, letters of intent and finding comparable buildings.
Token costs, valuation shortcuts, bias toward flattery and keeping a human at the helm.
Featured Guests
Meet the Hosts

Elizabeth Gearhart, Ph.D. is the host of Real AI Use Cases: Business Owners Roundtable, where she explores how businesses are actually putting AI to work. She is also CMO of Gearhart Law and co-host of the nationally syndicated Passage to Profit radio show and podcast.

Richard Gearhart, Esq. is co-host of Real AI Use Cases: Business Owners Roundtable and founder of Gearhart Law, a full-service intellectual property law firm. A longtime podcaster, he hosts Intellectual Property News and is the host of the nationally syndicated Passage to Profit radio show and podcast.
Elizabeth and Richard share how they value a business with Copilot, plan a book from the podcast and debate AI risk.
Episode Summary
Todd Drowlette, Brandon Bellamy and Sam Desany share how a commercial real estate team uses AI, from a site search that went from three weeks to under four hours to a four-agent system that writes website code. Richard Gearhart describes getting a back-of-the-envelope business valuation, and Elizabeth Gearhart announces a planned book from the podcast.
The roundtable then weighs the risks: token costs that can exceed the savings, companies rehiring people they let go, AI that tells you what you want to hear, and bad actors using AI. The group stresses keeping a human at the helm.
Frequently Asked Questions
How can AI speed up commercial real estate work?
What does it mean to use AI as a lens, not a crutch?
How does a multi-agent AI system work?
Can AI replace employees to save money?
Why is it risky to rely on AI for decisions?
What are the hosts planning to do with the podcast?
Episode Transcript
View Full Transcript
Now it is time for Real AI Use Cases: Business Owners Roundtable. I am going to pick on three of the people here and ask each how they are using AI in their business. They will give me one way, and then we will do a roundtable where we discuss AI. There is a lot to talk about these days with AI. I will start with Todd Drowlette. What is one way you are using AI?
I hired Sam, and Sam is running the AI for us.
And that is what a lot of companies do.
An example I can give, because he does a ton of things with AI, is that we used AI to streamline. I have one tenant called Label Shopper, Peter Harris Clothes. They are an off-price clothing chain, much like a TJ Maxx or Marshalls, and I work with them in 17 states. When we say, we are going to go out and we need six stores for back to school, we used to have to physically search the internet and pull up markets we want to be in. It would take three people full time for three weeks to identify potential locations. Using AI, Sam wrote a program that goes out, searches the internet, pulls a bunch of marketing brochures and runs demographic reports, all in real time. In a matter of three and a half hours it did what used to take three people three weeks full time. It will also auto-complete things like letters of intent and contracts. It is really good at that, so it saves a ton of time.
And those things are so tedious to do.
That is one of the things I trust AI to do. It is the tedious stuff, where you can just scan it and say, yeah, quick fix.
Let us go to Brandon Bellamy. What is one way you have used AI recently, or your favorite way?
I like to use AI to identify opportunities. I use it more as a lens than a crutch. Right now I am selling a building in Syracuse. For easy math, say that building is 19,000 square feet. I use AI, make a project or sub-chat, and upload the local assessment report. I have all of the tax maps for the whole city of Syracuse, so I run that report and find buildings that mimic the building I have, and I contact those owners to see if they are interested in another 19,000-square-foot building.
Wow, that is really smart. We get a lot of repeats on this show, and I have never heard that one before.
I wanted to backtrack to the sources, and me not using it as a crutch. The last thing I want is to have to be asked by a landlord or a client where I got information and not have a real-life source for it. I cannot just say, Claude popped this out, or ChatGPT popped this out. So again, I am big on using it as a lens and not a crutch.
Sam Desany, you are the AI guy. What is your favorite way to use AI?
My favorite way is probably to code some of the sites we are working on, the front end and back end. My background in school is a lot of marketing and data analytics. A lot of what I have Claude, Codex or Lovable doing is coding the back end and front end based on complex prompts I give it. It gives me feedback, and I can send it straight to the internet, or I can text Todd at 2:00 in the morning, which sometimes I do, with a finished product saying, how does this look?
Yes, it is doing the coding, but it is really your prompt that makes the difference.
The system I use is that I actually have four different AI agents. I have my brainstorming agent that I send my half-formed ideas to, and that agent talks to one that is already automatically connected. They create prompts that are sent to a third agent, and that agent gives me prompts and feedback and the specifics I need to send to my actual coding back-end and front-end agent.
Can you go over that again and bring it up a level? What you said is very smart and very technical, but our audience is very high level. People know nothing about AI, so try to explain it in more general terms.
Say there is a person who is coding, an employee, and you are a manager at this firm, but the coder only really speaks in coding language, so you cannot communicate with them. The manager has ideas, comes into the office and talks to a team of two or three other employees who act like a translator. They take the manager’s half-brought-to-life ideas.
Like all managers.
They take those ideas, digest them and send them to the coder in a way the coder can understand, so the coder does not have to deal with the back end and front end and with these half-put-together thoughts.
That is one of the important issues. We work with an AI company, and I am not always sure the person actually working on the project understands the context of how it is going to be used or why. If you have a way of communicating the bigger picture, I think that helps the coder and ultimately makes for a better project.
Yes. We are going to come back to you for a little more discussion, but Richard Gearhart, what is one way you are using AI?
One way, just this last week, was that I put our financials into Copilot and got a valuation for our business.
Is this your official announcement of your retirement?
No, not yet.
I am not letting him retire. He cannot sit around the house. I will never retire. Keep working.
I thought it came up with a fairly realistic number, one close to what I had, and gave a range. It also gave me suggestions on how to improve the valuation of the business, which I took to my business advisor, who immediately said, it is all wrong, but actually I thought some of it was pretty good.
I would use a second LLM and ask it, but I know you do not want to put anything through a non-enterprise version.
In the past you would have paid somebody maybe 15,000 dollars to do a valuation, and it would have taken two months. They would have interviewed everybody in the firm and done all sorts of things. If you are just looking for a back-of-the-envelope number, it works pretty well.
For me, I decided I want to take all the content from this podcast and have AI help me write a book, and Richard will be a co-author of course, because he is here gathering the content too. To post the podcast, I get the transcript and put it through ChatGPT, and I am going to automate this process. I am working on an agent for it. I say, ChatGPT, generate the description for Castos, where we upload our podcast and which distributes it everywhere, and make it SEO and AEO, optimized for search engine optimization and answer engine optimization, with FAQs, headers and so on. ChatGPT has all the transcripts already. The first step is to contact everybody and make sure they are okay with being in this book, and secondly I might put them in Gemini and say, give me a book, and see what it comes up with. It is still going to take a lot of human intervention and a lot of work.
You are just saying that because you want the copyright.
I asked ChatGPT what I should call this book, and it said to call it Real AI Use Cases so it matches the podcast. Then it said you need a tagline, and gave me a big long one. Richard and I talked about it, and Richard actually came up with what I thought was a very good tagline. It is going to be Real AI Use Cases: 50 Ways You Can Use Artificial Intelligence to Improve Your Business.
People love numbers.
They do. Spoiler alert, we might include 51 ways, a little extra value.
We are going to put 2026 Edition, because we want to try to do this every year. I feel like it is a documentary of how businesses have evolved using AI.
But if we are even still here in 2027, is AI not going to kill us all? There is a 10 percent chance that we are all going to be exterminated by AI.
I asked Gemini how it could kill us. It does not have hands or arms, so what is it going to shoot us with? It gave me a couple of things it could do to kill us that I do not really want to talk about here.
The fact that it had an answer is scary.
You have to put gaps in processes. You have to treat AI like you treat nuclear weapons, like the nuclear code.
It is called a human at the helm. That is the term you are looking for.
What do you think?
I think we are creating all these things and making our own worst enemy, but I think it is a tool that should be used. I do not think we should rely on it as heavily as we are. Even with your financial information, you want to be careful about what you put in there, because if your ChatGPT gets it wrong one day, it could expose everything you have going on.
That is the risk.
You look at these big companies, and quite recently, in the past couple of months, they are starting to hire people back that they let go because they had all of these AI and software systems. A year, maybe two years ago, everybody saw them letting all these software engineers go. Now they are hiring people back because they realize it is hard to manage and it actually costs a lot more money using AI, with tokens and credit usage for big things like Claude Code. You can easily spend 3,000 dollars a week on Claude Code with it just running 24/7, and then you ask yourself how much you are actually saving, whereas I could just have kept the person who was working for me before.
I saw that companies were not realizing the ROI they thought they would with AI, but that is a really good point. We talked to somebody who said you can have Claude do your video editing for you and tell Descript what to do, but you have to buy credits from Claude and from Descript.
And you still have to prompt it. It is not just thinking for itself and saying, here is a suggestion. You have to come up with an idea. It is not a person, and I think it is a great idea to use it as a tool. It is better used like that. If we have people who are operating it and who do not do stupid things like give it nuclear codes or people’s financial information, then you will be fine. Use it for tasks that are repetitive and that you do not need a human’s brain to do.
I think the danger from some bad actor using AI to commit atrocities is much higher than AI just trying to take over. One of the early AI cases I remember was in the biological sciences, where a research lab in Europe created over 50,000 compounds using AI. These compounds were toxic to human beings, and a lot of them were not even detectable using current medical tests. If somebody really wants to wreak havoc on the human race, they can just make a poison. I still think the probability of that happening, not that I am trying to give anybody ideas, is much higher than AI just deciding one day to exterminate the human race.
I do not necessarily agree. Even with my chats, sometimes the AI gives its own opinion, or tells me what it thinks I want to hear. Let us say you want to spend 100,000 dollars of the company valuation on a Bentley. I like Bentleys. You deserve a Bentley, right? But if your AI project manager or whatever chat you are using does not think that is a good idea and blocks your card, it is getting involved in your life in a very negative way.
Even for research it is iffy. I put together some questions you can ask ChatGPT to see if it recognizes your business and how your AEO marketing is doing. I tailored it for Gearhart Law, and I kept asking, and Gearhart Law kept coming up as an answer, which was great. Even though I said, assume I am a small business owner looking, I still think it is trying to give me positive strokes so I will keep using it.
The way you phrase the question has a ton to do with it, and then it will give you what you want to hear.
So I think you have to have somebody else ask it, someone you have never interacted with before.
Sometimes I find myself going to AI without even spending time thinking about the issue myself, so I have to watch myself. I do not want to lose my own ability to solve problems and think about things first, and then go through and verify with AI. It is dangerous. You can start going to AI for every little thing, and who knows what the consequences are long term.
It is still wrong a lot of times, so I use different LLMs and compare their answers. Sometimes they are aligned and sometimes they are not. I have my website on Manus. They had to take it down and put it back up, and it worked, so my website is back. I asked Claude whether Manus was operating normally, and Claude said it had been doing great for 90 days with no disruptions. I said, really? Having to pull all of its data off of Meta and then reestablish its data on its own servers in its own company is not a disruption? It said, oh, sorry, I did not look at that part of it.
It can be bad to go down a rabbit hole like that, which is another big reason I have those four agents before anything gets put into code, because in 99 percent of use cases they correct their own mistakes. Even if they do not, I verify the prompts before I send them in. It is more a faster way of doing what I am already doing. It is not creating something I do not know how to do.
For people who do not know what agents are, they will just do a series of tasks for you, right?
Agents can work in one chat at one time. Say you give a prompt. You can work from one agent doing one task from that prompt, or simultaneously you can work 100, 1,000 or 10,000 agents at one time. They are like employees.
But the thing to remember is that the most important agent you will ever have is your commercial real estate agent.
I was going to say your secret agent. Or your patent agent.
Exactly. With that, thank you very much, everybody. This has been Real AI Use Cases: Business Owners Roundtable.
Keep exploring the show
Hear how more business owners are putting AI to work in their day-to-day.
