Choosing an AI platform is becoming a long-term business decision. Once it knows your company’s data, workflows and institutional knowledge, how easy will it be to leave?
In this AI Companion episode of the Sure Oak Podcast, Nick Fraunfelder joins Michael Sengbusch and Suprit Patel to discuss the next phase of competition between OpenAI, Google and Anthropic. As models become cheaper and harder to distinguish, the conversation shifts to platforms, memory and the tools businesses increasingly depend on.
They also discuss where they’re investing in AI over the next six months, why productivity gains remain difficult to translate into measurable ROI and how autonomous agents are creating new security concerns.
00:00 Teaser
02:05 The significance of model distillation and resource savings
04:00 Security concerns and vulnerabilities in AI systems
07:55 The AI platform race: Google, Microsoft, and others
11:59 Predictions for AI in 2027 and potential risks
20:01 The impact of AI on healthcare and proprietary data
29:53 Future investment strategies in AI and industry outlook
39:55 Security breaches, shadow AI, and regulatory challenges
50:08 The evolving role of agents and AI in daily life
Please visit:
https://ninthhealth.com/
To contact Nick:
https://www.linkedin.com/in/nick-fraunfelder/
To contact Suprit:
https://www.linkedin.com/in/suprit-patel-a752914/
To contact Michael:
https://www.linkedin.com/in/msengbusch1/
To schedule a call with Sure Oak:
https://sureoak.com/strategy-call
Nick (00:00.000)
Welcome to the AI Companion podcast. It’s been a little bit since we’ve done this.
Nick (00:14.118)
Gentlemen, I just want to let you know President Trump did call me and ask me to change it to the Super Intelligence Companion podcast. And with us today are a bunch of the super intelligent human beings, Supre Patel, Michael Sangbush. How are you doing, gentlemen?
Mike S (00:19.363)
Superintelligence. Good.
Mike S (00:29.753)
to an awesome. Awesome. And as much as I’m sick of like changing the name of things, there is something to be said for not calling it artificial.
Nick (00:30.994)
Good.
Suprit (00:31.181)
Yeah, all good.
Nick (00:40.605)
Yeah.
Mike S (00:41.395)
there there’s there’s something there. I don’t know if we need a formal policy on it, but there’s something there.
Nick (00:47.590)
Is it better than the Gulf of America or is it or is it
Mike S (00:51.277)
Yeah. It it it we’re getting to the point where it is kinda super and not just artificial. Something there. Yeah.
Nick (00:59.868)
Yeah, yeah. I don’t know. I mean, you know, Jensen Huang is sitting there next to him and he wants to promote his super intelligent data centers. And so it makes sense from a brainy perspective, like it makes America great. So I don’t. I mean, I’ve I’ve I’ve come to be very cynical now in in my older age. So if you are interested in sticking around and listening to the whole podcast, we have a number of topics we’re talking about today. turns out it’s no big deal to build a frontier model.
Suprit (01:08.824)
Yeah.
Mike S (01:12.122)
no.
Nick (01:29.663)
the new AI race is the race for a platform. What is what are Google and Microsoft going to respond to the two IPOs, or I guess three IPOs now, in the AI space. And then what we are going to do is tell you how we are going to invest in our businesses in AI over the next six months. And we are going to give you the sure to be wrong, entirely way too early predictions for 2027. So first I will throw it over to you, Suprit.
let’s talk about Kitty. they made a new model. Cheaper, faster, smaller. yes, it was distilled. Yes, it was all these things. But hey, I made a model, no big deal.
Suprit (02:11.150)
I mean, the distillation is I think gets all the headlines, right? It but it really saved them time, if anything. I think of some of the news stories that are out there and what you think about it is really it’s really about the amount of resources that they’ve been able to throw behind it, right? Like experts. Like if you if you look at AI, it’s it’s beyond just a piece of technology. It takes it takes clinicians, it takes lawyers, it takes accountants, it takes all these professions to to come up with real tasks.
Nick (02:30.002)
Yeah.
Suprit (02:40.577)
Right, to be able to train it and make it useful. And I think that some of these open source models, especially when you’re when you’re coming out of China, you you have the labor pool, you have the talent pool, to to really dedicate to to that to that task. And I think it’s gonna it’s gonna sh reveal itself in in a matter of months where they really catch up, people start adopting, and it it’s coming, right? But it I think it’s distillation is that easy clickbait piece that everybody likes to jump on, but I think there’s a lot more behind.
Nick (03:09.992)
Do you think we should explain what distillation is? I don’t know, Michael. Do you want to take a shot at it? Or I I can’t. I can take a crack at it. Whatever you what do you want? You well, so the big model is the teacher. So in this case, in Kimmy, the the teacher model was Chat GPT, was I forget what version, it was the latest one. And the student model was Kimmy. It was a smaller model. And so therefore, it’s just asking prompt after prompt training the smaller, cheaper model.
Mike S (03:14.177)
No, let’s redo
Do it.
Suprit (03:17.785)
Yeah, yeah, go ahead.
Nick (03:37.970)
what what the actual right response is. Now I’m sure there’s way more that goes into distillation, but just think about the scaled over millions and millions and millions of prompts training this model. And there you’re there you have it. So there there’s a lot of problems. It’s not smarter. It’s not better than the old model. It’s just cheaper and smaller. So
Suprit (03:53.357)
Yeah. And you’re able to catch up quicker, right? So somebody get they’re they’re able to catch up to a baseline and now they can kind of really jump off of
Nick (04:00.732)
Yeah. Which for us as consumers a huge win ’cause token cost.
Suprit (04:04.675)
Yeah. Yeah.
Mike S (04:05.719)
Yeah, you’re getting the you can see the token costs are coming down. And I think I saw an interesting graph this week where it was the average token costs coming down. And then it also showed I think it was Anthropoc’s or OpenAI’s revenue flatlining at the same time. And it was almost like directly correlated on how the graphs looked. And that seems like what we’re seeing play out very publicly right now.
Nick (04:14.994)
Good.
Nick (04:21.500)
Mm.
Nick (04:33.746)
Mm.
Mike S (04:34.773)
ahead of IPOs, which is exactly what’s that you know, is it an existential threat? Is it an existential threat to humanity, or is it really an existential threat to these businesses as they prepare for an IPO? And that’s kind of where this is at. And if you can have cheaper models or you’re just using the last generation models with no problem from a personal use standpoint and then from a business use standpoint,
Nick (04:48.050)
Yeah, yeah.
Mike S (05:03.726)
You’re getting to the fact where there you’re getting to the point where there’s some diminishing returns on the latest models. You’ve seen this last year, the year before, there was the problems of like hallucinations, and you were concerned about what the AI was gonna say. And now we’re at to a point where we’re more concerned about what the agents are gonna do. And we’re kind of into this next phase of this.
Nick (05:07.666)
Yes.
Nick (05:27.944)
Do you think the models, I mean, w with diminishing return. Are the models, I mean, for every man, every woman, are are they at where they need to be? I mean, what how much more frontier do we have to be? I mean, I I don’t I don’t really I can’t tell the difference anymore. And the switching costs between models, no big deal to me anymore. Maybe a couple things.
Mike S (05:47.063)
No, and the switching this the switching costs are easy. And the analogy that I think holds here, you know, a little bit, at least on the consumer side is side, is it’s like flat screen TVs. Like they work and they’re cheap and there’s each generation is not meaningfully better to justify a much higher price point. but that technology is going to continue to improve and on the enterprise side or special use cases, and especially as we get more onto the agentic side of it, those
Frontier models will be important and they will be valuable, but we are well outside of the consumer use case and we’re well outside of the regular business use case. I don’t know what you guys think about that.
Nick (06:27.580)
Yeah. Yeah. Yeah.
Suprit (06:30.412)
Yeah, I mean I I think they’re out kicking their coverage by a lot. Cause I mean the the fact is most people are not using this. That’s just, you know, we we always feel I think three of us or people in our circles who who tend to, you know, use these kind of tools fairly often, it’s easy to think everybody else is using them. And they’re just not. The adoption has is not it’s exactly. It’s not widespread. So you you you roll these new models out.
Mike S (06:53.529)
They’re using it for glorified search.
Suprit (06:59.818)
And and the capabilities are awesome. I mean, no doubt, right? But it’s a narrower and narrower and narrower audience that can take advantage of what’s being put out. And the risk goes up and up and up and up. So you’re like, you’re coming down in the number of people you can actually serve and and can benefit from this new, you know, impressive resource. But then but then the risk of the security side, the safety side, and all the agents running amok and everything like that is just going.
you know, in the other direction. So I think that there’s obviously been some coverage about, you know, mainstream usage. It’s not where it it needs to be or where people think it is. And it’s also, like you said, great point. It it’s it’s widespread, but it’s not it but it’s really shallow. It it it really is like search. People are just typing into it as if it was Google, right? Back in the day. So I I feel like it’s you’re rolling out this new tech, your user base shrinks as you as you roll it out, the risks go up.
And then it plays into the whole, you know, the revenue side, the monetization side, everything and how does that
Nick (08:03.452)
Yeah. Michael, I was talking to Saprete last night on the phone and he goes, Holy crap, man. I you know, he’s been in his his head’s been down on his new startup. And so he he comes back and I send him the news stories for today. And goes, Holy crap, I I I I gotta catch back up on what’s going on in the world. But I felt the exact same way. This whole concept of continuous agents, like it just caught me off guard. I didn’t I I’ve never used dots, I’ve never used Muse Muse, Grokbot. Have you ever heard of Spark?
Suprit (08:27.661)
Yeah.
Nick (08:32.230)
Or anti gravity? Have you heard of these products that are out?
Mike S (08:36.075)
No, but I I have heard good reviews on Muse and Grokbot and Dots obviously came out like last week or something. So that’s that’s a pretty new one there. But that there’s something to be said here for everything we just said around the commoditization, the commer the commercial use case, the consumer use case of this, kind of devolving just to glorified search, maybe some image mock-up stuff that you know people use pretty regularly.
Nick (08:39.859)
Yeah.
Nick (08:46.216)
Okay.
Mike S (09:04.558)
There is something to be said that if these type of agents work out, like a Muse, which is packaged, I’ve seen demos of it. I haven’t used it, but it’s packaged in a way that makes the functionality that open claw
Nick (09:20.243)
Mm-hmm.
Mike S (09:22.411)
shocked the world with kind of you know four or five months ago. it packages it in a friendly, easier to use way that’s baked into the security and kind of the platform that Meta supports. You could we could start to see that kick off next, where now you’re getting to the agent side, it can actually do things for you rather than just something that you talk to, which that use case seems like is getting pretty standardized now. You can talk to a
Suprit (09:48.898)
Yeah.
Mike S (09:49.665)
It’s a glorified search. You can do some image, you can do some mock-ups, you can do research, you can create documents, you can have projects. That’s kind of known. But if this nice, kind of friendly agent works in a way that’s not scary for people, consumers may move into that next. But that’s this is like of the last week or two kind of stuff.
Suprit (10:11.970)
Yeah. Yeah. Yeah. No, I think I think what you’re kind of hitting on is that there’s opportunities to kind of unlock new customer behavior, right? So I think it was very similar to what what OpenAI did with Chat GPT, right? They they they rolled this out and it got mass adoption and you unlock this new behavior and and you people started using it and then it opened up the world for Claude and all these other tools to say, you know, this chat interface, which maybe
Nick (10:12.060)
Yeah. And we don’t know what the adoption’s gonna be. Sorry, go ahead, Spree.
Suprit (10:42.121)
a company may not have thought would be the be all end all or some major part of their their strategy became a really big part of it, right? So I think that Muse is it and dots and all these platforms, are they gonna hit on something that customers want and didn’t realize they wanted and it was just simpler than we all thought.
Mike S (11:00.590)
Yeah, and I was kind of using the this whole it kind of just devolved into glorified search as kind of a pejorative there. But the way you just said that, Sapreet, is actually makes me think about it because if if you are if that type of behavior, glorified search, is replacing Google, that’s huge. Because I mean Google is the dominant search engine, right? Like no one uses
Nick (11:23.186)
Mm.
Nick (11:27.303)
It is Sir J.
Mike S (11:28.607)
It is, it’s like it’s synonymous with search. You know, it’s not there’s no duopoly here. There’s a monopoly on search. It’s Google. And if a product comes in and people are now using using it as glorified search, we we really minimize that. I didn’t realize that until the way you just kind of described it, where it’s like that’s a really big thing. Like that’s a massive change for consumers.
Nick (11:35.432)
We do.
Suprit (11:50.261)
Yeah.
Nick (11:51.479)
The consumer behavior portion of that is so interesting. So I mean, like we, you know, one of the things we talked about was this A16Z article about open AI understands something important and rare, which is a platform play. and the idea that the idea that they made the chat, like that was the innovation. Everyone, all the scientists says, Well, yeah, we’ll just do this little product, but let’s get back to super intelligence AGI, let’s get back to the real work. And it turns out
Suprit (12:11.222)
Yeah.
Mike S (12:15.991)
Mm-hmm.
Nick (12:18.930)
I think it was Peter Thiel who calls Sam Almond and goes, No, no, no, no, no. That is the product, the chat GPT, that is the product. And now they’ve moved into like these other things. It’s really interesting how consumer behavior is very unpredictable. So back to your point, Michael, about are you gonna replace Google search? Well, chat GPT has made Google search human. I mean, we know there’s a lot of robotic or or agentic search going up, but human search has gone up ever since the advent of people chatting things. And like
Suprit (12:46.123)
Mm-hmm.
Nick (12:49.126)
I didn’t expect that. I thought my business would be over. Like I’m post I’m supposed to be dead as an agency. But there but no, it’s actually it’s hand in glove. These two concepts of SEO and GEO. They’re so related and human behavior is unpredictable.
Mike S (12:59.266)
Mm-hmm. W one one feeds the other. Yeah. They they feed the other. It’s a huge, it’s a huge change on that. but then this go back to this whole who comes up with the platform.
Nick (13:13.192)
There we go.
Mike S (13:14.051)
Where is the moat? And is the moat the applications? Is it the ecosystem? Is it the memory?
And the portability of your memory from a lot of these things. I mean, one of the things I was thinking about in terms of switching, because we said it at the beginning of the podcast, but the context switching between different AIs is fairly simple. I have two running at all times. And I use one for business, I use one for personal. And it’s one’s Gemini and one’s Claude. It’s this isn’t it’s not radical. And if I’m in X, I’m using Grok just because it’s there and it’s connected to the stuff. So that type of context switching is pretty simple to do.
Nick (13:41.116)
Yeah.
Nick (13:48.924)
Yep. Right.
Mike S (13:53.751)
What’s not simple to do is replicating the work that you do in one of those channels in another one. Like I gave you pretty clear use cases, business versus personal. And there’s probably dozens of others that you’re kind of smaller use cases. But if a year or so into this, the memory and the training and how it knows my stuff, that does start to become emote, starts to become sticky, which would make it hard to move.
Suprit (14:00.695)
Mm-hmm.
Nick (14:23.814)
I mean, what do you think about what op I have you s played with their docs and their PowerPoint and all their tools that they’re building and
Mike S (14:31.651)
Great. Supreme, what have you seen?
Suprit (14:33.935)
I’ve not played with the those pieces of it, right? I mean, I think it’s clear why they want to go in that direction in terms of like owning a stack, right? So if you’re if you’re already in OpenAI or Claud or whi, whichever one you choose, and you’re you’re running your your your queries through there, you’re doing your research through there, yeah. I think it makes sense for them to deploy their own version of slides or PowerPoint or Word or, you know, whatever it is. So I I think that that’s a natural thing, but I do think, you know, Michael, I
I think piggybacking off what you said, I’ve been using mostly Claude for for the last, you know, number of months, if not, you know, a year. I’m not switching. Like it would take it would take a lot for me to switch. I’ve gotten really comfortable. I don’t see a big change in UI in terms of how I interact with the application. They’re the same. I I put a big premium on memory and context. And I think now that this system has
Mike S (15:21.719)
The UIs are the same too. The U is the UIs are the same for everybody. Yeah.
Suprit (15:32.115)
learned what I’m trying to do in my various projects, the thought of retraining is it’s just it’s off the table, right? The only thing I would consider is in the future potentially moving to an open sourced model where I would feel more comfortable plugging in data or personal things that I don’t want the Frontier Labs to have. But I I’m not I just don’t see people going from one to the other.
Mike S (15:49.059)
Mm-hmm.
Mike S (15:54.711)
Yeah. And businesses that have invested in one and were invested into anthropic and into Claude have that problem or that scenario times a hundred. Where it it there’s there is a enterprise knowledge and enterprise memory that really helps. And I’ve found it it really does start to know how you work, what you’re looking for, and I’m not correcting it as much as I used to.
And it has a general understanding of our business. Where, like, I don’t know if you guys I don’t know if this is just me perceiving this or if this is really kind of what’s happened, but I used to have to always like frame up questions a lot more, like take this role or act as this or research this or make sure you know these things first, and then we’re gonna do now. I’m kind of finding like I don’t need to do any of that stuff. It just
Suprit (16:40.554)
Mm-hmm. Right.
Mike S (16:51.371)
Remember this isn’t a business context. It’s like it knows those things already.
Suprit (16:55.766)
I think that it’s starting to incorporate context and reasoning in a way that we it it required time for it to build that capability, right? Because you need that memory, you need that the the repeated use to kind of build that. So I do think that’s there. Now, where I can see things not fracturing but but opening up a little bit is when you get to privacy concerns, right? So if you’re if you’re a company and you’re
Are you really going to plug in all your company’s financial data, all your company’s personnel data, or you know, things like that to a Frontier lab and have it totally accessible there because you know, you know there are breaches happening all the time. And I think that’s where there’s going to be a change. I think that’s when people are gonna start shifting and having hybrid approaches to their their kind of AI infrastructure in a company, right? Where you run certain things private, almost like the equivalent of like on prem.
And then you run other things like, you know, through through the larger models in in the in the cloud, so to speak, in that analogy.
Nick (17:59.238)
Yeah. Yeah. That’s really good. I mean, I just you know, I think about open AI, the they’re like dunking on Gemini right now. We got what they got codecs, they they got they got the coding, they got the stuff coming, and they’re and and you remember and everyone’s just like, Well, Gemini like is it’s unusable, has no has no good harness, has no of this. Do you remember like back in like when we’re like,
Google nailed it with with 3.5 flash. I mean, like, they just crushed it. They’re gonna be the dominant. That was only six months ago, seven months ago. And then it’s like these leads, like we’re talking about like, open AI is gonna dominate the world. It’s gonna be great. Like, Google can’t do what open AI is doing with these constant releases, like, yeah, we got a slice, we all this. Yeah, but Google has to like worry about what they have, like a million products or something like that.
Suprit (18:37.707)
Mm-hmm.
Mike S (18:54.082)
No, and I and I had this, I remember these discussions from like six months ago or so when we were having this. I pushed internally at at the business Gemini quite a bit due to its integration to G Suite. And I started using it a lot and then finally got clawed. And then I was like, yeah, like, okay. Gemini’s fine for some QA, it’s fine for
Nick (19:08.968)
Right.
Suprit (19:09.088)
Mm-hmm.
Mike S (19:20.269)
Some light research obviously has access to search a lot better. But for the business stuff that I’m doing, not only is Claude better, but Claude has continues to get better. And then I just toggle between which model I’m using based on the task in order to manage my co my token costs. I don’t know if you guys are doing that. Are you just sticking it on Opus? I never use Fable. I haven’t had a I haven’t had a use case that warranted Fable. I’ve never even done it because I don’t think anything
Nick (19:21.725)
Right.
Nick (19:34.760)
Good.
Suprit (19:35.306)
Yeah. Yeah.
Mike S (19:48.993)
in the in Opus isn’t getting it done for me. I don’t what you guys think.
Suprit (19:52.148)
Yeah. No, I totally agree. Totally.
Nick (19:55.603)
Yeah, ’cause you you you built your whole company on the the l the the first iteration, right, Sapreet on on Opus? Yeah.
Suprit (20:00.683)
Yeah, all on in Claude, absolutely. And never had to call on the latest models. I mean, I I just it it goes to what we talked about earlier, where the the the user base for those models is just very s it’s small. I mean, I don’t the the workload that you’d have to have to need a a model that size, not to mention the cost if you’re a startup, right? it it’s just it’s you know, it’s just too much. And I do think, Michael, to your earlier point, like this Google thing
We’ve talked about it in the past of they have the biggest distribution engine there is, right? With G Suite, with the business, and still they’re not getting the the penetration and growth rate that they would want on a lot of these new tools. And I’m I’m not saying they’re dead or it’s not gonna work out for them. There, they’re clearly they there’s gotta be a plan. But right now, Claude works perfectly fine with.
With Google Slides, I mean open like you can do what you need and still operate both environments.
Mike S (20:58.775)
it works.
but one of my things that I think Claude has done the best on is understanding not just how to help you organize thoughts, but yeah, th this might sound Mickey Mouse, but making a slide that you don’t have to go edit too much kind of like it it dials it in now. Like it knows style guides, it kind of knows what we’re doing. I have a f like a flow on how I manage that stuff and that stuff helps a lot.
Nick (21:15.838)
Yeah.
Suprit (21:17.010)
Absolutely. Absolutely.
Nick (21:27.356)
Yeah, no. Yeah, once it gets to know you. I think maybe that’s where the switching cost is. I mean, I was about to call BS on you, Supreme, because you did switch from open AI to Anthropic, right? Like Yeah. Because I think like if I’m stepping back, I’m trying to be reasonable about this. And if my mic sucks, I’m sorry, I didn’t know it turned off. if I’m trying to like what the experts say is anthropic only has a six month lead.
Suprit (21:36.170)
That’s true. That’s fair, fair.
Nick (21:53.843)
Technol like technology is all they have and like everything that they’ve done, whether it’s from UI or personalization, you know, they have to maximize the six month leave. That’s why they’re so in bed with the government. You know, that’s why they’re trying to create their own lit legislative mode. Because they know this isn’t gonna last for them, because it will be copied. Eventually, Google will do something. We but like I mean Microsoft is nowhere to be found in this conversation, and they’re the also the other huge platform with huge distribution.
Suprit (22:07.254)
Holy.
Nick (22:23.912)
But have no Apple was the other one, no solution.
Suprit (22:26.983)
Apple, yeah. All the massive distribution layers aren’t aren’t aren’t, you know, taking advantage of their user base. Yeah.
Mike S (22:28.162)
None.
Nick (22:31.784)
Google’s the only one. It’s shocking. It’s shocking. So all right. So next six months, what are you guys investing like? Okay, so you’re doubling down. You’re all you’re all doubling down on anthropic. You’re drinking the IPO Kool-Aid. You’re gonna, you’re gonna buy some shares. You you’re gonna get a five X ROI on your on your investment. What what what were you guys investing? What what it even for your businesses?
Mike S (22:55.800)
Yeah. Yeah, I think it’s kind of similar to the last time we talked about it. We we are in a create an ROI story. And we do get pressure on what’s the formula, how are we gonna do that? The way I think about it right now is plenty of anecdotal stories around projects that would have taken X amount of time were done in half amount of time or whatever it might be. We have plenty of those. adoption is
uneven in the business. So we monitor that. Costs are under control. We don’t have any weird spikes. I think we have a good line of sight into how those token costs are going to increase, kind of plateau, really. So we feel like we have that on lock, which is great. Justifying that spend above and beyond anecdotal use cases is a challenge. We get enough questions about it that we have to do a little bit more than let me show you some anecdotes.
on this, but then there’s a little bit of a are you deferring costs, are you deferring hiring because of AI? And we’re not, we’re not like we’re we’re not cutting staff because of AI.
Nick (24:04.188)
Right.
Mike S (24:12.803)
But there’s probably a do you need to hire more people story? Where are we getting more productive? Are we absorbing any type of employee departures in a way which doesn’t require a backfill? Like is attrition just something that does help you save costs and you can absorb them better because of AI? I don’t have hard data to show that, but I have a
Nick (24:18.387)
Mm-hmm.
Nick (24:31.070)
Mm.
Mike S (24:42.081)
An inkling that that might be true.
Nick (24:44.678)
Interesting. Sprite. What about you?
Suprit (24:45.876)
No.
I mean, for us, you know, we’re early stage. So as we build out our prototype and as we actually build out our first version of our product and and really take it out to market, you know, I I’m more concerned with how do we build something that isn’t easily replicated with through an AI engine, right? And so that’s more a focused thing on like what kind of proprietary data can we capture? How do we integrate into workflow?
Nick (25:14.845)
Yeah.
Suprit (25:17.439)
How do we how do we establish like a trust layer, especially in healthcare, right? Between a patient and and a system or a patient and a provider. And how do we bring in clinicians into this loop so that it’s not just something that you could put into Claude and get done? So for us, it’s about how do you build something that’s totally proprietary and how do you take advantage of the fact that I do think in healthcare, especially, maybe other industries like finance, we still have a little bit further to go.
In terms of establishing trust, patients aren’t just gonna tomorrow let an agent prescribe medicine for them. Right. So there is a trust layer that I think we benefit from in the healthcare world. And so I think anytime we can, you know, integrate a physician in or a clinician in or make like real human decisions when it comes to that point, and then do that quickly, right? And fit that into a person’s day and make it less heavy lifting for them.
Nick (25:54.311)
Mm. The agent doesn’t want to.
Suprit (26:13.652)
the better off we are in terms of creating something of value that you can’t just go have an agent replicate quickly. So that’s that’s kind of our focus is how do we make it really proprietary, you know, bring in the real world elements of it and and really grow through that that lens.
Nick (26:15.890)
You bring us in.
Nick (26:20.326)
Assuming.
Mike S (26:30.273)
Yeah. And that touches on what I was referring to before is more of, you know, how are we using AI internally? And what I left out was is what types of AI are we introducing into the product itself? And we are making a lot of advancements on that. And when we think about AI that manifests itself into the product, some of that AI is backend. We do a lot of integrations and we are deploying agentic agents to help with our integrations.
Nick (26:30.705)
Yeah, because
Suprit (26:42.483)
Mm-hmm. Okay.
Nick (26:42.841)
Mm scared.
Mike S (26:58.116)
So when things fail, can it self-heal? We’re doing some really cool things there. And then more on the customer-facing side of the product, we’re dipping our toes in the voice agent water a little bit. and we’re doing some really cool things with conversational BI. So, how do we enhance our analytics product by being able to ask questions about it? Can you put an AI layer on top of your data? So
Suprit (27:14.047)
Mm-hmm.
Mike S (27:21.709)
Those things are like are in active development and many of those are either making the product more stable or they’re making the product more valuable because we’re creating new products out.
Nick (27:32.095)
That’s really cool. I mean, like to your point, Sapreet, like how do I use the tools to actually differentiate? And like no one thought about using AI in this way. I mean, like, that that’s that that’s really that’s the cool part about AI. for sure, we’re actually on a little bit more of a negative cycle on AI. It’s just like it’s just very the trust is at an all-time low and security risk is at an all-time high. And
Suprit (27:46.527)
Yeah.
Mike S (27:50.988)
You think?
I a little bit.
Nick (28:01.034)
Shadow AI was never, I mean, like we always tolerated shadow AI in our organization. And now it’s like a, now it’s like, whoa, the shadow AI is really powerful. Like your ability to take your own personal AI and start like, if like there’s instances where like it would request access into our environments. So like, what is this AI? It’s like, it turns out it’s Joe down the street, you know, next door who wants to use his, you didn’t even know he did it.
Right. And so these are the things that keep us up at night now as the security and agents. And you know, people are like, Well, can I do this? Cause I have I built this whole thing on my AI that knows it’s just like we were saying before. I have my my personal Claude over here. It does all these things and it’s been helping with me on my job. It’s what developers have been doing forever. I’ve been doing my job coding for ages and not letting you know it. You know, checking in code that wasn’t mine. And while I’ve been going to the
Suprit (28:41.919)
Yeah.
Nick (28:57.286)
Starbucks and sipping lattes. That is happening everywhere. So that that is something that we’re on the cycle of just very mass security. but still building our tools, trying to make our lives better, implementing. So all right, entirely too early, sure to be wrong, twenty twenty seven predictions. Who’s who’s got one?
Suprit (29:19.689)
I mean, I I think the most obvious one, it’s not even a prediction, right? It’s probab it’s already come true, is the level of security breaches that are gonna it’s coming is going to be massive. Because it’s not just rogue agents and the hugging face and all that, right? I think that now that you have potentially an avenue to mass adoption of agents, like through Muse and through other platforms, and if you get people
Nick (29:37.469)
Nick (37:39.821)
Yes.
Suprit (29:48.713)
Connecting those to email, financial records, like whatever it is, I mean, these are all to me, and and I’m not like a security expert, but I I feel like these are all entry points, right? And I think that I think something big is gonna happen. And it’s it’s gonna be fairly disastrous, not from like a human casualty perspective, but from an infrastructure perspective, potentially, right? I think that if you look at, you know, water, electricity, you look at our power grids, you look at
financial institutions, there are there’s too many vulnerabilities in this ecosystem and we’re we’re far outpacing the their ability to defend with what we’re developing.
Mike S (30:30.885)
Yeah, you touched on two things there, and I think these are these guys I had written down some predictions. I think that rogue agents and managing agent traffic first could do what Sabreed said. There I think there’s some looming disaster at some point in the next twelve to twenty four months.
How that’s created or how that happens or what causes that would be up for debate. But I do think there’s something looming. On the agent side of it, one of the things that I think is interesting coming up is because so many agents are doing so much work on the web or through web apps.
Mike S (31:13.262)
Do agents start getting throttled? Do they start getting blocked? And do you start having do we start to have to create some type of a credentialing system with agents that sites are already doing this? I think Amazon blocked Muse.
Nick (31:26.994)
Mm-hmm.
Suprit (31:27.284)
Mm-hmm.
Mike S (31:28.555)
So great. Okay. You can just kind of do a a full blown, like a block. You could block all agents. You could, you could do that. Or what you start to do is start to say, actually, cert agents are good. We do want certain. And therefore, there has to be some almost like an API authorization. There’s some type of an encryption or some type of a certification or credentialing that allows your agent to interact with that app. Because we do a lot. Yeah.
Suprit (31:43.262)
Yeah.
Suprit (31:55.976)
Like a KYC.
Nick (31:57.639)
It’s a KYCF. Financial institutions have had this, yeah.
Mike S (31:58.751)
Okay, we’ll see. Yeah. Yep. That’s gonna ha I think that’s gonna have to happen pretty quickly. And we’re kind of in that, you know, if you if you go back four or five years, there’s the wild, wild west of how the AIs were created. They just crawled the web kind of anonymously and gobbled up all the information, ignoring copyrights. And then it was like, hey, we built something, it’s too late to undo it. So there’s that kind of like wild, wild west piece.
Suprit (32:15.934)
Mm-hmm.
Mike S (32:26.935)
And now you’re in this wild, wild west of agents that can just do stuff. And you gotta think that at some point those things get blocked or constrained, or there’s gotta be some something in there that happens. And I and I don’t know if that happens first or if what Sabrita was talking about happens first, where like
Nick (32:36.678)
Mm.
Nick (32:46.536)
Someone wakes up.
Mike S (32:47.972)
Someone wakes up, like, do you need like, and I and I don’t, and I don’t use this analogy lightly, like, does there need to like a 9-11 AI problem where now we go, crap, we have to really start monitoring and lock things down? I don’t know, but I do think there’s looming things out there. And specifically, Sabrit, since you and I are in healthcare, what would happen if there was some type of a healthcare breach that then took a lot of our customers?
and really elevated their suspicion and concern around do you use AI? Because we already deal with that now when we’re talking to customers around where we use AI, how we use AI, and how we protect their data. It’s it’s already ratcheted up pretty high on what we can and can’t do. And if there was a healthcare breach, it would just like nix it for a long time.
Suprit (33:30.888)
Yeah.
Suprit (33:36.808)
Yeah. And it could be it could be a data breach, right? A healthcare entity, you know, personal information, all that stuff. Or it could be disruptions to the supply chain. I mean, there’s there’s there’s more than one way to for these ro for these agents to get out of control and and wreak havoc. And I feel like now that we’re talking about it, it’s it’s like you’re unleashing this technology this really advanced technology onto the public that just is not prepared on how to safely
Nick (34:05.022)
Don’t know how.
Suprit (34:06.121)
deploy it and use it. It’s it’s very similar to when all these companies started going to the cloud, they started hosting things in AWS and Azure and and then controls on those environments were left to people who maybe didn’t set the right controls, right? Inside of those environments. And then you have, then you have breaches. And I feel like unfortunately now we’re talking about every single person who has a phone could download Muse, right? Like anybody can download Muse. And you you’ve just opened it.
Nick (34:07.814)
They don’t know.
Nick (34:32.488)
Yeah. Yeah. Working agent. I don’t know. I love all the healthcare guys. Always worried about all these you know, after the Experian breach and the change healthcare breach, everybody who’s listening, your stuff’s already out there. You’ve been breach like your stuff like who cares if more of these things get breached? I don’t like it’s not gonna matter. All you healthcare companies go for it. I don’t think it matters at this point. There’s nothing
Suprit (34:45.106)
Your stuff’s already there.
Mike S (34:49.857)
Yeah, the yeah.
Suprit (34:53.225)
Yeah.
Mike S (34:54.989)
Yeah, and the health Well, they they it’s funny you mentioned this ’cause they I think they were talking about this on on all in podcasts this week or last week, ’cause you know, Chamath was at Facebook.
Mike S (35:18.214)
Nick (43:20.566)
Nick (35:30.120)
Mm-hmm.
Mike S (35:37.069)
You would have these scenarios where, like, because go back to like ad targeting stuff, you would start to get like there was a couple really prominent ones where, you know, like because healthcare information was used in Facebook ad targeting, women or girls were getting retargeted with like pregnancy stuff. There was a couple high profile ones like that.
And then parents are looking at it and it’s like, why are you getting targeted with this? And it’s like, it’s because your daughter’s pregnant. There was like a couple examples of stuff like that that happened. And that’s because healthcare data was leaked into the audience targeting. And then Facebook knew that person was in that audience and they targeted them for some type of a healthcare product. And by doing that, you are letting other people know about somebody’s health status. And you could see the same thing happening with healthcare data in AI, where now
Nick (36:14.237)
Yeah.
Nick (36:25.714)
Yeah.
Nick (36:32.616)
Yeah.
Mike S (36:33.773)
Does an AI start to know about your personal health care? And does that get leaked to other people? You could see the same exact thing happen.
Suprit (36:43.315)
Mm-hmm.
Nick (36:43.570)
that’s so good. That’s really good. I don’t really have a prediction, but it is something I would like the community to solve is how easily now in my in my community is the problem, the SEO community, is how easy it is to manipulate answers. the data is, you know, I have this guy named Mark Williamson Cook, really smart SEO practitioner. And
You know, he started this whole thing about like what’s the best way to get brand mentions or you know, and he he like, well, you gotta download this cat.txt file. And like that was all of a sudden now, now it was like, hey, when Google’s saying, like, well, how do I get brand mentions? Like, well, you gotta download this file and you gotta do this thing. And it was like it was completely a hoax, right? And so we at we’re at this point of where it’s garbage in, garbage out,
We’re starting to load up the internet with really bad data whereas training. So I would love to see in twenty twenty seven, you know, maybe better uses of AI than agents. we’re always trying to put travel agents out of business. The internet forever is trying to put travel agents out of business, right? That’s the first thing we’re gonna do. Online travel agencies, agentic travel booking, that’s like great. I’m glad.
But can we actually like use it where there’s actually validation layer with math and science and zero data training and actually solve some real problems in the world with AI instead of forcing agents down my throat? I just I don’t
Mike S (38:11.405)
Well, the th there’s part of that like forcing agents down people’s throats where there is there is kind of a prevailing wisdom when you talk to a lot of the you know, a lot of the founders of the Frontier Labs, a lot of the big tech companies of at some point does the UI completely become agent based and agentic based? And I think we talked about it on a previous podcast. Like, I don’t know, like parts of it maybe, but like people still are gonna want to go to the web.
Nick (38:30.310)
Jetic.
Suprit (38:31.390)
Mm-hmm.
Mike S (38:41.687)
And they’re still gonna wanna use that web app. It’s not gonna be totally just some
bot some agent that’s just gonna do stuff for you. Like I don’t think we’re there yet. I don’t I don’t know if that ever gets that way, but there’s trying to jam it down a free threat. Yeah.
Nick (38:55.450)
I Amazon didn’t I mean Amazon put a lot of malls, but like it changed the whole shopping experience. I still want to go and have life experiences, but like I don’t want to buy widgets and commoditized stuff anymore at a store. I want it sent to my house, right? So we’re gonna see a lot more of that.
Mike S (39:09.966)
Well, like with Amazon too, like you could you could have an Amazon agent that does shopping for you. I’m sure they probably already have that or somebody’s built that. But
Nick (39:16.028)
Yeah. Yeah, just make sure make sure the air filters are here on time. You know, I don’t care about much necessarily.
Mike S (39:19.831)
Yeah, but you’re right. There is there right, but you’re st you still wanna look at it. You still wanna browse it. You still I mean you might not be going to the mall to do it, but you’re gonna look at it. There’s the the the user interface is not going away and it’s not gonna become some, you know, automated, you know, whatever in the sky. I d I don’t know. I don’t think we’re there yet.
Nick (39:24.766)
Sure.
Nick (39:38.077)
Yeah. I also am curious about in twenty twenty-seven these coding layers, like this AI. will it I mean, cause Google’s not really making a bet on one of these harnesses. They’re just gonna say eventually it’s just gonna be a part of Gemini. You’ll just speak it into existence and on the platform itself, all the back end, all the database, all that stuff will just be chosen for you and it’ll just be you’ll just be able to start making video games like we always said you could.
Mike S (40:05.239)
Well on that note, go ahead, Sabrina.
Suprit (40:05.520)
Which No, I was gonna I was gonna say, I mean, which is why I think if your value prop is that you’re building something on top of the model, it you it doesn’t and and you’re entirely dependent on that model to do like the deterministic, you know, reasoning or whatever it is, then your company’s not gonna be around for that long, right? I mean, that’s where like when when we go back to like what we’re focusing on over the next 12 months or or the way we’re gonna build value is
through proprietary things like data, you know, clinical inference, like thing things like that that are outside of the model. Because I think Google’s bet is that this just gets swallowed up. Like you don’t need that that tool over time. And yeah, maybe they missed the boat on, you know, some of the the coding harnesses, but eventually the model’s gonna do it. So you know, what did they do? They they maybe just missed the boat for six months or twelve months.
Nick (40:58.470)
Yeah, I mean their bad is like, we got the best coders in the world. We’ll figure it out.
Suprit (41:02.119)
Yeah.
Mike S (41:02.926)
Well, you know, speaking of coding, and let’s kind of getting into like more 2027 predictions, and I think we’re already seeing some of this now. But what I’m gonna be curious about is a lot of the code that was created in 2026 that was done with a lot of agents, and a lot of lines of code were created. What is the long term support and stability of the of that? Because what I’m hearing, and I’ve seen it internally, but at our company.
Suprit (41:26.449)
Yeah.
Nick (41:26.536)
Yeah, yeah.
Mike S (41:33.035)
And I’ve heard friends talk about it and I’ve heard different versions of the story, but you’re getting certain features and functions out the door, you’re building new products. Now, what happens when you don’t understand really what that what was in that code? And you do need to go in and start to fix it. Now you’re left with, do I have more agents going in to fix that code? And if I have more and more agents building the code and then fixing the code and then testing the code, you are getting further and further and further removed.
Suprit (41:48.008)
Yeah.
Yeah.
Mike S (42:02.455)
And what happens is they create many lines of code, oftentimes unnecessarily. So you have sometimes redundantly because the agent is doing code and it doesn’t know about the rest of the platform. So it kind of rebuilt stuff. So you may be seeing the output of new features and functions in the product, but that underlying code is becoming
Suprit (42:03.859)
Yeah.
Mike S (42:25.545)
another black box and you are further abstracted from it and it might work in year one, but as year one turns into year two and year five, what exactly how do we go back through that and see w exactly what was built?
Suprit (42:35.538)
Yeah. I think that’s a great point because you know, as you we’ve unleashed this army of people who aren’t coders and who are now using these tools to code, I think what’s gotten compressed or maybe even yeah, same even eliminated is in like this whole SDLC, like the testing, like the QA, the UA, like all of those pieces we don’t really hear about, right? What we hear about is just the deployment, deploy, deploy, deploy, not.
Nick (42:37.233)
Mm.
Suprit (43:03.494)
You know, the testing environments, the dev and like it it is it is a a concern. I think the way you put it highlights that perfectly. I it it’s just where the attention is has just shifted.
Nick (43:15.398)
And it’s not two thousand five anymore. We’re talking about a very mature tech environment that we’ve unleashed these these programs on.
Mike S (43:15.619)
You’re gonna see
Suprit (43:22.394)
And we’re using APIs and other integrations to be you know, to call it all external services and so your your vulnerabilities are
Mike S (43:26.903)
Yeah, and it’s gonna get it’s a weird it’s gonna get so complicated on what was built that you’re gonna have to further rely on more agents to go in and tell you what the other ones did because you won’t be able to go through it. Because A, you didn’t build it and B, it’s probably more than you would have created. I think that’s gonna be a real issue. It’s almost like your classic Gartner hype cycle where there’s this kind of peak of inflated expectations and you’re gonna get into this trough of disillusionment for a while.
Nick (43:51.836)
Yeah, yeah.
Mike S (43:54.819)
Where everybody’s gonna really second guess that. And I think you’re already hearing some of that. Where like six months ago, there was like, I built this, and it’s like, you know, it was vibe coded that, or our team did this and generated 10 times more code and blah, blah, blah, blah, blah. And now I think your people are kind of reacting to that, where they’re like, I’m not really sure that actually accomplished what I needed to do. And now it’s a maintenance nightmare. And you’re that’s gonna, that’s gonna happen in 2027.
Suprit (43:58.641)
Yeah.
Suprit (44:19.312)
Yeah.
Nick (44:20.574)
Well, you have to pay premium for a grok super intelligent architect agent to come in and read your code and give you a recommendation about how to fix your crap. Because there’ll be an agent for that, right? And okay, well, here now you have to hire my sub agents and my coder to agents, and everything’s gonna be narrow case use agent. man, I don’t I don’t I’m not I’m not looking forward to that. All right.
Suprit (44:21.990)
Yeah.
Mike S (44:32.226)
Yeah. I think there’s gonna be some of that. Yeah.
Suprit (44:35.495)
Yeah.
Mike S (44:39.532)
huh.
Mike S (44:44.024)
Yeah.
Suprit (44:44.166)
Yeah.
Suprit (44:47.746)
Yeah.
Mike S (44:48.090)
There’s there’s there’s a story I tell all the time when I think about a lot of a lot of what non-technical founders do is non-technical founders have some product they need to build. And regardless of if they outsource it or if they hire like a junior person to do it, as soon as they hire that senior technical lead, I said, I don’t care how good that outsourced job was or how good that junior person does.
Whoever you hire to be your senior technical lead will tell you that whatever that guy or this guy built was crap. And I was like, 99% of the time, that’s what’s gonna happen. And with these agents, you’re gonna kind of see the same thing. Like, yeah, I made this real quick, I made this real quick. And then another agent’s gonna come in. They’re gonna be like, Yeah, I can go fix that for you. But what that other agent did was a piece of crap. I have to re-architect it for you. And you’re like,
Nick (45:23.513)
Because that was their job.
Nick (45:41.702)
Well yeah, because it’s gonna be made in the image of the creator, right? Like well first, like we’re all we’re only gonna make money you’re only gonna make money if we inspect the roof and you find a hole.
Mike S (45:44.865)
Yeah, it’s gonna be some other like I wouldn’t have done it that way. Yeah, it’s gonna be arguing what’s
Yeah, it’s like you can pay me more tokens because I’ll fix what the other guy did with your tokens from last year. I’ll come in and have to fix it, and it’s gonna be it’s what’s gonna happen.
Nick (46:00.294)
It’s just human nature. First thing a CMO does is build a new website. I mean it’s like, yeah, this is what we do, right? So I hope agents I yeah, and agents if if it’s making money proving their worth, they’ll do the same thing too.
Suprit (46:05.840)
Yeah.
Suprit (46:11.728)
I’m not a real big fan of the agent personification that we’re doing here. Like we’re making them real people. Like next thing you know, there’s gonna be like an HR department for these agents to go hash out their, you know, their their concerns against each other. I it’s a scary future that we’re going down.
Nick (46:19.518)
Unfortunately.
Nick (46:23.410)
Blade Runner, yeah.
Nick (46:29.244)
Yeah, agent rights. Let’s start it right now. This podcast. Gentlemen, it’s been fun. we’ll do it again probably in a month or so and we’ll yeah, we’ll catch up on stuff we don’t know very much about. We’re this are our predictions. Thank you, gentlemen. Any any big news or things we need to announce? Supre. Go braves. Yeah, big game today, right? yeah. yeah, they’re already in. They threw the wild card. Sorry. Missed that.
Suprit (46:30.844)
There you go. There you go.
Mike S (46:40.408)
Forward to it.
Suprit (46:42.576)
Yeah. Sounds good.
Mike S (46:46.593)
So yes. Go go braves.
Suprit (46:51.078)
Go Braves. Big win. Big win last night.
Mike S (46:53.249)
No, they won yesterday. Dodgers tomorrow.
Suprit (46:58.406)
We finally got the Phillies off our back.
Nick (46:58.652)
I’m a big fan. Yeah. Yeah. Sabrita, what’s the new company?
Suprit (47:02.352)
go. Yeah, yeah, no, I’m I’m heads down. You know, we’re we’re launching we started Ninth Health this year and it’s it’s been going great. We have early clients we have pilots that are getting ready to get kicked off we got financial commitments from a couple of funds we’re we’re really rocking and rolling so it’s it’s it’s moving fast and I think it’s a testament to this this whole industry and the technology that’s available to us and and the use of it and kind of responsible use of it.
Nick (47:14.546)
What’s the elevator pitch?
Suprit (47:32.144)
I’ll say. but it’s it’s going great. And I look forward to updating you guys on the next pod. Yeah, we’ll go with go deeper on
Nick (47:37.480)
Cool. We’ll drop it in the show notes. So give give me the website. All right. Talk to you guys. See ya.
Suprit (47:40.816)
Yeah. Thanks, man.
See, bye.
Mike S (47:43.715)
You guys.
| Cookie | Duration | Description |
|---|---|---|
| cookielawinfo-checkbox-analytics | 11 months | This cookie is set by GDPR Cookie Consent plugin. The cookie is used to store the user consent for the cookies in the category "Analytics". |
| cookielawinfo-checkbox-functional | 11 months | The cookie is set by GDPR cookie consent to record the user consent for the cookies in the category "Functional". |
| cookielawinfo-checkbox-necessary | 11 months | This cookie is set by GDPR Cookie Consent plugin. The cookies is used to store the user consent for the cookies in the category "Necessary". |
| cookielawinfo-checkbox-others | 11 months | This cookie is set by GDPR Cookie Consent plugin. The cookie is used to store the user consent for the cookies in the category "Other. |
| cookielawinfo-checkbox-performance | 11 months | This cookie is set by GDPR Cookie Consent plugin. The cookie is used to store the user consent for the cookies in the category "Performance". |
| viewed_cookie_policy | 11 months | The cookie is set by the GDPR Cookie Consent plugin and is used to store whether or not user has consented to the use of cookies. It does not store any personal data. |