The conversation between Danny and Jim O'Halloran was informal and exploratory, covering several AI-related and business topics but producing no concrete decisions or next steps. It began with a brief exchange confirming that time had been set up for Danny to meet Bob the following week, and with Jim saying he would simply thank the person who arranged it. The two then looked into a strange, unformatted block of text that appeared in an email thread; Danny explained it was the full thread combination from Bob, Jim, and Amy, likely a rendering artifact from different email providers (Yahoo, Outlook, and Gmail) rather than anything sensitive.
Jim then asked whether Bob understood AI basics like agents and bots. Danny explained that "basics" depends on the era a person learned about AI, and described Bob as likely a sophisticated technical person who may have trained an open-source language model on his own company data, though that comes with risks because AI hardware and software evolve quickly. Danny walked through how he explains agents—instructions plus skills (like SOPs) plus tools and channels, all built on top of a large language model—and noted that definitions vary widely across the field, so meeting people where they are matters more than jargon.
The discussion then turned to proprietary data protection. Jim asked how Danny protects a client's information as an outside consultant, especially given free AI plans that train on user data. Danny said he isolates client work in a specific Tailscale-type VPN mesh network, restricts his tool list to those guaranteeing zero data retention and no training, prefers local small language models that never reach the internet where security is a priority, and would encourage clients like Bob to audit existing systems and press their vendors on the same guarantees. He also explained that over 80% of his recent usage has shifted to Chinese AI labs like DeepSeek, which he says are more performant, more economical, and tunable, with inference running locally in southeastern Ohio.
Jim asked what would happen to Danny's clients if Danny were incapacitated or died. Danny showed his client portal, which stores full transcripts, outputs, and artifacts of every conversation, and described his philosophy as building systems clients can eventually own and run themselves—modeled loosely on Hinge's "meant to be uninstalled" pitch—with the goal of leaving clients humming rather than dependent. He also demonstrated the live transcription app capturing the current call ("OHC Transcript"), described his in-person setup using a recording device, diarization, a screen facing the meeting participants, and a thermal receipt printer with plastic sleeves that produces a one-page summary and action items before clients leave the meeting—though he would not bring that setup to a first call with Bob, just paper.
Finally, Jim asked about cash flow management for Danny's business. Danny said he has a financial dashboard that manages contracts and proposals, that he's financially able to invest in himself without landing contracts for a year, and that he and Ryan may put something together instead of using their current accountant. He explained the Ascendance contract—billing software for telematics, expanding to Allegiance, contracted since 2024 through Plum One Development Partners—is billed annually at $36,000, is up for renewal in the next couple of weeks, and that after this cycle the goal is an acquisition discussion. Danny also mentioned using virtual cards for online payments since 2021 to avoid fraud and unwanted subscription charges. Jim shared background on Bob's businesses, including building locations near the Iowa State Fair grounds, his daughter's charity work with something Jim couldn't name, and Bob's involvement in affordable housing projects in West Des Moines. Danny said he maps a person's entire business ecosystem, timeline, industry trends, and known investments before meeting them, spending around $75 on pre-meeting research. The call ended warmly, with Jim wishing Danny luck with the Bob meeting and Danny saying he'd let Jim know how it goes.
Answer you that you have a.
She was answering you that her and I have already set up time for me and Bob to meet next week. Well, actually, I'm meeting her, too, I think. Yeah.
Okay. So I'm just going to thank her for doing that. Okay. And. And did you see the. The email when I scrolled down because I was trying to see if you were responding and has all kinds of things of back and forth and. Do you still have it? Do you still have that?
Oh, I definitely have. Oh, let's see.
all the. Yeah, let me see. Anyway, so it goes down. It follows everything. Danny just came to me. Maybe it was just to remind you of an appointment. Is this right? Have an item. Oh, I see. That's where you are. But do you see that? It's kind of funny how it's. Listen, maybe it's just mine.
Yeah, I don't know because I see the. The whole thread, but.
Do you see the thread? It's really weird. I mean, it's that it looks like something you maybe wouldn't want to share. I do. Is that coming from her or coming from.
Are you talking about specifically like this. This whole block here, this kind of unformatted block of notes?
Yeah, yeah, yeah, yeah.
Yeah, there's nothing. There's nothing we can break it down here. Let's just.
but why. Why would it even show up? Is it from her computer, do you think?
Every email platform is different. We're talking in here, we have Yahoo, Outlook and Gmail across multiple people. So I'm guessing, I'm guessing it's simply a one of, you know, the translation from one inbox to another to another. Because here's the breakdown of those. I mean, it's a combination of threads. It's. It starts with Bob to you saying, hey, it was good to see you, and then Bob, and you replied to Bob and then you followed up with Amy. So that's what I'm. All I did was take this whole blob and I broke it out into the individual messages. So there's nothing nefarious or. It's all the thread.
Okay, because I didn't see it just seems like if she were sending this and all this came in, it would be probably something I don't know if she'd want.
this is. I think this was just like a weird thing with us all having different email providers. Honestly,
Okay. Okay. So here's, here's my short questions for you. When the first night, the first time we met Bob at Lyceum, I think it was the first night I think you mentioned for sure. The second night you mentioned bots and agents and I think, I think maybe even the first time and then, then I asked you something and you said I have agents working for me all the time. Okay. Right now. So did you think that Bob understood, does he understand some of the basics of AI about agents and bots and things that did you get some sense from him or is he just doing some investigation right now?
So I would say that the basics of AI are going to depend on the era you're reflecting on. You know, this goes well back into 1940s of some of the more like significant research in AI and so a lot. So folks coming from, you know, who are aware of AI, if you were aware of AI before, you know, 1990, before 2000, before 2010, before 2020, each of those are vintages of people that clearly went out of their way to try to like understand this thing that didn't have value yet at least couldn't have the articulable value that most people were thinking. And you're thinking about events like, I don't know if you remember the IBM, the, the computer that beat the chess master, you know, and.
Right. So those are the moments that broke through the zeitgeist. So when I think of basics, I think it really matters of like, what era did you pay attention to this? Because also one's attention burns out on the complexity of this after I imagine, unless you're crazy like me, a number of years. And so for Bob, what it sounded like he came from was understanding and being a tech technical person like in, in these past decades. And he also then indicated things like the way he framed it kind of comes two ways to me. He says we have our own large language model. I didn't realize we had an Iowa based Frontier Lab. And we don't. And so it's kind of the tongue in cheek saying, like, you have your own large language model. Like you must have invested every dollar you've ever made to get something that even gives you a half decent output. Or what he's really saying is, I've taken one of the open source language models and I'm taking my data and I'm training that model open source so we can make it my own to be useful specifically to my companies. That is the bucket I'm guessing he's in, which still requires a lot of sophistication to even know that's an option and know
Yeah,
how to start. And you also need a lot of capital and you have to probably be a little ADHD and curious to like throw a lot of capital at something like, could we have our own model? So that sense, that's what I got, the sense I got there. I also got the sense that he set it with a plum, as, as in he's had that model for 12, 18 months maybe, at least. Right. And what an awful situation to
have invested a ton of money into a point in time on this technological ramp. When the innovation curve, both in hardware and software that serves AI, has grown so fast that if I say I'm investing in System A today in six months S, that could have been an awful bet because now we fundamentally cracked the way that XYZ works and now compute costs are down 20%, which means XYZ is available, by the way. Models have moved forward, so he could be in that mess too, where they have invested in something. And it's not really punching, you know, it's not like a computer where the hardware was the hardware, it performed at a certain rate, it was going to live for a period of time. Like this is more like get down. It gets down to time economics. If I could give you a model today that performs 5 to 10x the speed of Bob's model at the same cost, you would choose that every time, right? Bob doesn't have that choice if he has a bunch of hardware and server infrastructure where he's like hosting and running this model. I hope he's not, because the cheap and best fitting method. If you are in that bucket where it's like, I know my business, my business has proprietary data and that data could work really well if an agent get into that concept, which, the concept of an agent very overblown. The core is still the large language model. The agent is simply, you know, like on your computer you have your file system. So the, the best way I've heard agents explained is by this company called Vercel. And this is how I bring it. This is literally what I bring up for everybody when they ask me, what's an agent? The general structure of an agent is the idea of a directory is a series of instructions. One set of those instructions might be, you are a helpful assistant. Your role is to find things that the user might miss and generate helpful nudges throughout their day. That could be the whole instruction file, just a simple file. Here's this example right here of like the agent directory. So there's instructions and then there might be skills you can think about skills like SOPs. When we do billing, how does that look? What steps do we take? What systems are touched? That same structure is exactly what agents are expecting. They're expecting a skill so that when you say, go check my email, it doesn't just go to your email inbox and delete everything, or go to your email inbox and do nothing. Because you made an email skill and you put it in its directory just in a folder, it could first say, okay, I've been given a task to go read emails. Let me check my toolbox. Oh, there's an email skill. Okay, this is how I do that. Does that make sense? Like
Yeah.