Show notes
Scot Morris, OD, and Rehan Ahmed, MD, on how fast-moving AI is shaking up traditional software models, healthcare tech and beyond.
The episode opens with a company that used AI to reverse-engineer Adobe’s suite of creative tools and build a free, open-source alternative called “Photograph” in two weeks, a sign of how easily AI can now challenge large software companies and subscription models.
Dr. Ahmed then unpacks the term “harness,” using the metaphor of a horse: just as a physical harness directs a horse’s raw power, an AI harness gives an underlying large language model direction, memory, rules and guardrails. An AI agent = model + harness.
To close, the hosts debate whether per-seat software subscriptions are disappearing as small teams gain the ability to build, tweak or copy complex software in days, what that means for EHRs, and why revenue may shift toward data ownership, data sharing and compute consumption.
Transcript
Welcome to Real Talk on AI in Eye Care with your host, Dr. Scot Morris. And I'm co-host, Dr. Rehan Ahmed. In each episode, we discuss current news, add in a little AI education, and debate innovative AI tools and topics that will change the industry. All right, welcome everyone to another episode. Scott, I know you have some interesting news of the week for us, this episode. Yeah, I do. And you know, I don't think this for the audience, this isn't necessarily something that deals with eye care directly, though it's going to deal with the technology that you've all heard of. I don't know if any of you guys saw this this week, is that most of you've heard of Adobe, and maybe even many of you have seen Adobe or use Adobe Creative Suite.
You know, and then that's basically been rebranded as Creative Cloud, and we use it in our office all the time, and it's kind of a really high subscription fee. It's got a lot of services, and that's great, but basically they kind of own the market on creative content development over the last couple of years. Well, then last week, a new interdict, a new software developed called Artcraft. In fact, it's a new free open source software suite that looks very, very similar to all of Adobe's editing tools. And it's called PhotoCraft. Now, the interesting thing about this is that it's not a secret. This company basically developed an algorithm to reverse engineer Adobe's big seven tools using the newest version of Cloud 5.5.
And they were amazed. They did all of this. They recreated what Adobe took years to do in two weeks. They basically said, Adobe, we're done with you. And they created a software that does everything Adobe does, or almost everything that Adobe does, for free. And I think, Rayon, you know, what that makes me wonder is, are we going to see more of this? Have we seen the end of subscription services? And you and I talked about this a couple weeks ago and said, you know, hey, sometimes we get sick of being subscription to death. And this is the first really, really, really big company that I hate to say it, but maybe lost everything in a weekend. And so I want to save that. I want to talk about your knowledge, because I think it's going to be a good lead into some other things we're going to talk about.
But I want to come back about this in our fireside and talk about, are we starting to experience the end of software?
Yeah, I saw that news. And then the first thing I did was look at the stocks of these companies. And they definitely did take a hit. And for all of us that deal in software, we sort of think about these things. So, yeah, what for our knowledge, it's actually related a bit to what you talked about, Scott. It's a term that has been coming up more and more, and the term is harness. It's all over my feed. And for those who follow AI, I'm sure they've been seeing this word pop up. But I just wanted to spend a little bit of time unpacking what this means. So I'll sort of define it here in a second, but just think about what an actual harness is, right?
Scott, you know what harness is, because you have actual harness. I mean, you have horses. Oh, yeah, I have horses. Of course I got horses. A horse is basically, you know, and think about horsepower and all this. So horses, for us who don't have horses who live in, you know, not in places where I can have a horse in my yard. Yeah. A horse is a huge amount of power. A horse power all these things. A horse standing in a field doesn't plow anything. It doesn't pull anything. Just a horse on its own. You need to have it, quote unquote, connected to a harness to power up for work. Scott, I could see you rolling your eyes. This is a metaphor that you're like, seriously.
But it points a horse in a direction, and it keeps it from running off. That's the point of a harness. It generally, yeah, that's exactly right. Yeah. And so AI borrow this word. And, you know, I think it's actually a decent fit. And so when you use like a large language model, chat, GPT, collage, MNI, on its own, it does one thing. You give a text, it gives you text back, and that's the sort of simplest way you can use it. On its own, on its own, pull up a patient. It doesn't remember necessarily what you told it yesterday unless you sort of log in stuff. I mean, you have to have it connected. It may not even know what day it is until you sort of plug it in.
And so you see this when you AI search the web or you put notes in charts. It's something else. When there's memory involved, when it's actually doing actions, it's something that, and what the thing is doing, it's actually invoking a harness. So let's suppose you have an AI that's sort of working through like Wacoma follow ups, and it says, you know, I need the last three patient OCTs. The harness checks it. And what the harness does, it gives the model guardrails. Okay, am I allowed to see these? Do I have access? It goes and gets them a handsome back. And then it will say, okay, now I need also the visual field to provide proper context. Then it goes, gets a visual field and does that. So the harness decides whether to go straight in the chart.
Does it sit there until the doctor sign off? So here's what it asks. It acts, it looks, and then it asks again, right? And this perpetual loop over and over. And this is sort of what we talk about in AI agents, right? But what the harness does, it brings sort of what a harness really does, Scott. It brings the hands, the memory, and the rules and guardrails. So here's the depth. So we've talked about agents a ton. So the agent equals the model plus the harness. So the harness provides direction. It provides guardrails to the model, the underlying model, and that equals the agent. So you're going to hear a lot about harnesses and agents, but there's a subtle difference.
So harness is the, so the harness plus the model equals the agent. And so that's something I've been seeing this term over and over again. I thought I'd spend just a few minutes clarifying to the audience what that means. And I hope that's helpful. But just think about a real harness and it won't be a led astray. I love that. It won't be led astray. Well, I think that, you know, we, and I kind of, I want to extend that word harness, right? And I kind of wonder, the last couple of weeks we've been talking about all this stuff that's happening with, you know, AI getting out of the box and what it can do. And, you know, you and I have asked this question offline a couple of times.
I said, is the real problem that humans haven't put guardrails on? We haven't put a harness on AI really to control what it does for us. And I think that many of us out there, and there are many in the audience that you probably don't know enough about using AI that you really are worried about it yet. And then there's a group of people who are saying, let's see what it can really do and aren't putting any harnesses on it. Yeah. And then I think there's a group of people who are maybe the more the good actors, let's say, who are like, hey, maybe we need to control this. And I think that lends into kind of what I was talking about before. I think this Adobe, you know, this change that's happening because of our craft.
I think we're going to see that a lot more in the near future. And I often wonder, Rehan, we've talked, as I said before, we've talked about how subscription services drive us crazy, you know, blah, blah, blah. But, you know, I wonder, are we at the beginning of the end of SaaS services? And if you think about my thing about this is right now, the talk is they're aiming at Microsoft Office, right? And Google did that a couple of years ago or, you know, a decade ago. And so we're going to try to build a Google version of Office. And now we have sheets and, you know, all that kind of stuff. But now I wonder, you know, if you look at these companies that have spent, oh, my gosh, thousands of engineering hours and have those really crazy, complex code bases.
And, you know, the vendors are locked in. And that's why we paid these monthly, you know, per seat subscription fees, blah, blah, blah, blah, blah. But now when you have a solo developer, somebody in a garage or just a small team of people could say, hey, I can reverse engineer. 80% of this good stuff. I can get a subscription, reverse engineer it and create something for free or sell ads on it, make money that way. You know, this justification that we've always said of, hey, you know, yeah, it took you a lot to develop, but we'll pay a fee. I kind of wonder, is that going to erode away? And is that all going to, I mean, is that going to go away?
Are the moats, it seems to be a word that we've been using a lot lately. Are the moats the protective mechanisms for software? Are they now over? And what does that mean? I can't say as, you know, somebody, I look at all these different subscription services I have on a monthly basis. You get your credit card statements and you're like, oh my gosh, look how much I'm paying for all these subscription services. And then I'm wondering like, do I really need all these? Are there technologies out there that would do the same thing for less? And the answer is, yes, there is. I mean, I'd say we change what creative software we use, what AI platforms we use to create content.
I mean, almost every month we're using something different because there's something better, newer, faster, cheaper. You know, and I wonder at some point if just software will be free. And what does that mean? You know, as you and I talk about is we're both are developing software companies. Is that how, what does that look like? I mean, where's the revenue source or will there be a revenue from software or will the revenue source be totally just the data you own? And you and I've talked about that for the last two or three years is that I think we've been missing the boat in healthcare is that the money is in the data. It's not in the software, the processing unit of the compute.
It's in the data and, you know, I don't know. What are your thoughts on that? Yeah, I mean, I think definitely there's no there's no IP moat. There's no moat in software anymore. And we know we've seen the example that you've talked about. I think we've seen this for a while coming as the AI in terms of what it was, what it has been and what continues to be really, really, really good at its coding. Because that's what so there's just so much training data already available. That's how it became very good. And I think what people have sort of sounded alarm on are like on other jobs, right? Where the training data is now just getting more and more integrated.
It may get as good as those within software development. But let's just since we're talking about software, let's stick on software. I totally agree. I mean, I think software is quickly becoming a type of commodity. Let me give you an example, right? So I was a couple weeks ago, I was looking for I was looking for an app on a phone that could measure the pupil size with it after taking a picture of someone's face of their of their of their eye, their pupil. And I just want to like super simple app. I just want to know is it three millimeters, two millimeters, whatever, just doing that. Nothing too complicated. And I kept on picking a search for it.
I just got getting apps for interpupillary distance. There's some apps I download them. This is kind of confusing. Why is there a paywall? I'm getting ads. It didn't really work that well. So I was like, you know what? Forget this. Let me just go on Gemma. I want Claude. I could have gotten anywhere else. And I literally just built an app to do this in like 15 minutes. Not even. And I still use it. It's really good. I use it for all sorts of various projects I'm working on. And I mean, it's just my creativity that stopped, but I didn't have to look. I almost thought it was easier. It's a crazy world. It was easier for me to create a bespoke solution that I wanted myself rather than spending another 20 minutes of Googling for other solutions.
I think that's a feature. We've sort of talked about AI making your own movies, right? Oh, for sure. This is sort of the same thing. And software, I can imagine a world that, yeah, you're copying Adobe. That's great. But like, why not? Maybe you have a small tweak that you want Adobe. Maybe there's something better than Adobe. Yeah, there's lots of stuff I wish worked better. Yeah. Or maybe the specific use case that you want. And rather than having some contractor work it out for you, which would take tens of thousands of dollars or millions of dollars, you make your own software. I get it. And so in terms of healthcare, I mean, let's talk about EMRs.
This is something people pay six, you know, small products are paying an arm and a leg for. So this is something that's very interesting. And I wonder how they think about it. I do too, you know what? And it makes me wonder, you know, I've been thinking about this a lot the last couple of weeks, you know, are we shifting from, oh, for lack of a better term, software subscription mentality, our world being software subscription, software subscription, which really it has been for the last, you know, five, seven, nine years. I mean, since COVID, it's really, you don't own anything. You rent it, right? You have a subscription to software. And now I wonder.
You subscribe to your EMR. I assume you should, I have a subscription to my EMR. It's a SaaS solution, I think. Yeah, I do. I'm trying to change that, you know, by developing a better tool. But, you know, then I wonder, are we really moving from subscription service? So what's the new modality going to be like? Is it going to be compute? Are we going to be, or instead of using software, are we really just going to say, I'm going to pay for the amount of compute? And that's the tokenization and all that kind of stuff. And I go, okay, that's great. But, you know, we, I think this is a whole other real talk we cover at some point. What's the reality of these data centers?
You know, their lifespan is very short and very expensive. And what do we say? I think we talked about a couple weeks ago, right? They're going to spend, they're going to spend billions to make millions. Well, dot math doesn't work. And so, you know, I'm not sure that we're going from software subscriptions to compute subscriptions, but I do wonder, are we going to go based on consumption of data? Not consumption of compute, but consumption of data? Yeah, I think that's going to be an interesting switch on it. And yeah, I'm really in terms of our electronic medical record systems and how they're thinking about it. They have to be really thinking about the data and the network effects, meaning it becomes more powerful when not just one person using EMR, but 10 or a hundred or thousands.
And so they have to really become creative and really think about how it benefits me to use a specific vendor's solution. How is it making my data more relevant to me and how is it helping use other people's data to make it more relevant to me? Or will we really need to have all these separate EHRs? Will there be something that develops that says, nope, EHR is just a, it's just a bridge to a aggregation of knowledge that, you know, I don't know. I wonder this all the time. I mean, I, you know, I pick on EHRs quite often and say, I think that, you know, that there needs to be some changes, but they are kind of the backbone, but they're a backbone that doesn't have any vertebra to it, right?
Is you don't have any date, you can't have data sharing. And I think that's going to be a big piece of the future. And, you know, I wonder about, as we see this Adobe and lowering the, you know, getting rid of the moats and changes from subscription services to something else, whatever that's going to be called, I, you know, I wonder if your whole model is based on subscription fees, how long is it before you become the target of someone saying, no, I can do it better, faster, cheaper, and I can reverse engineer your stuff in a week or two. You know, I think about that all the time. I see technology and I'm like, oh, we can, my business partner and I, we can build this.
In fact, we can build it better, faster. And we've done some of that. Sometimes we've taken certain systems out there in the eye care space and said, we can reverse engineer and in a weekend, we can create something better, faster than anything that exists. And so I do wonder, you know, if I had a crystal ball and could look five years in advance and know will any of these softwares even exist, then I'm not sure they're going to. Yeah. Things get flipped on their head. I mean, SaaS is such a high margin business. And now we're talking about a essentially commoditized, no margin. I mean, things are going to be driven down to very close to, I think what you said, yeah, you're paying for compute.
You're paying for the core usage of what you're doing. So yeah, I don't know how this is going to play out. I mean, I did see some stuff that some of these free tools are buggy and they don't work perfectly. They don't work like the obviously real Adobe. But they were created in a weekend. Yeah, that's the thing. I'm sure. We'll figure out how to fix the bugs. You know, open source thousands of AI agents plus a couple of engineers probably can go through those bugs, I think fairly quickly. Can fix it. Yeah, I mean, this, you know, doing my business partner calls it bug smashing, you know, I mean, you can get on and bug smash with most of these things with AI pretty quickly.
As soon as you find a problem, go fix this. And within a few minutes, they can either build or pull code from somewhere else and fix it. So yeah, it's, you know, this is going to be one of those great podcasts for us to go back and listen to five years from now and go, will we write or were we wrong? Or was it somewhere in between? I don't know.
Well, everybody, I want to say thank you for the, you know, for listening to us kind of ramble on about what the future might look like. You can get, we have some really interesting innovator talks or podcasts coming up. So please, you can catch those as well as our weekly real talks. We have some guests coming on the, on our show for that as well as all the articles that are going into AIinicare.com, our journal site. And you can always reach us. Rehan, R-E-H-A-N, or Scott, S-C-O-T, at AIinicare.ai, that's the AI part. We always love for ideas, suggestions, and feedback about what we're doing and what we can do to make this better. Everybody have a great day.