Subscribe
Share
Share
Embed
Hosted by Runpoint Partners’ founders Sam Gaddis (tech entrepreneur & AI builder) and Matthew Hall (PE operator & growth strategist), Runpoint Podcast strips the hype from artificial intelligence and shows you how to turn it into concrete business results—fast.
Matthew (00:00)
Hello and welcome to the Run Point podcast. I'm your host Matthew Hall. This is Sam Gaddis. Sam Gaddis, happy fucking New Year.
Sam Gaddis (00:07)
Happy New Year, Matthew. Glad to be here. I got my Santa mug and I'm ready to go.
Matthew (00:11)
Good. All right. So this is our 2025 recap and 2026 predictions show. We've we've been in business now since April of 2025. We have it's only actually what nine months, but it feels like we've been around for longer than that, Sam, but we've learned some things. We're going to talk about what we learned in 2025, what we saw, what the trends were, and then make some predictions on what we think is coming around the pike in in 2026. How's that sound?
Sam Gaddis (00:38)
That sounds great. Let's do it.
Matthew (00:39)
All right, without further ado, let's jump in. I want to hear from you on in your mind, who was the biggest winner in AI in 2025?
Sam Gaddis (00:47)
Yeah, great question. think the biggest winners were the companies that didn't get discouraged or distracted by the many failures of AI. Of course, we're big boosters. We always talk about how great things are. But the truth is, going into January, AI didn't really work all that well. I think those of us who were really into it could see a growing exponential curve, especially in terms of rate of progress. And so we all knew that
what was coming as real and not a fad. But the models themselves, I think we were using 4.0 still at the beginning of the year, were not performant. They didn't work. You couldn't trust them. They didn't give you reliable results. This is when people were, people still talk about hallucination, but a lot less. At the beginning of the year, people were talking about hallucination constantly and a large percentage of businesses just didn't believe that it could do anything.
It wasn't until the middle of the year when we had GPT-5 and then at the end of the year with Opus 4.5 that businesses really started to be able to use these models in production tools, whether they be external or internal tools. So I think the, but we still see a bunch of people who are discouraged because they tried AI at beginning of the year before.
and walked away with a conclusion that this just doesn't work, it's not reliable, you can't trust it. see, like lawyers are in that category, I feel like. But the people who are the winners are the ones who retained curiosity and the ones who continued experimenting because it's working really
Matthew (02:16)
Yeah. Yeah. It's crazy to think Claude code came out in February, right? And like, in a real way, our entire business is based off of it, you know, and like, if that hadn't happened, I don't think we'd be here. You know, it's like, that was to me, that's the, that's the turning point for a lot of this. my, I took this a little bit differently, I think a little bit more macro. ⁓ and so my biggest winner, I was just trying to look monetarily.
Sam Gaddis (02:32)
What's yours?
Matthew (02:41)
at who's actually making money in AI right now, because obviously you've got the huge CapEx spend, you know, the boom. So Nvidia is making money and data centers are making money. People that help build those things are making money. But I think that's a little bit played out because, you know, after them, you've got everybody building frontier models. You've got the people who are spending the CapEx and they certainly are not making money. They have incredible traction, but they are hemorrhaging money, you know, and all of the training and all of the people.
And then you've got the other side of the whole market at the bottom, builders. So people who are building with AI, small teams. And so my big winner of 2025 are acquihires, are people in teams that have been acquired by large, comfortable corporations and have gotten themselves rich and have gotten themselves fantastic jobs. Those are the people who are making out like bandits right now. You know, we just saw last week, Manus get acquired for 2 billion, I think, from Facebook.
You know, know cursors, I think what are they, they're, they're now making a billion dollars, think, currently, you know, it's like that they haven't been acquired yet. You got things like windsurf, which is a cursor clone that I don't know anybody who ever used, who got acquired for a large sum of money by Google. And I think is now the backbone of anti-gravity. There are a lot of these small teams that built something cool with AI who have built themselves a reputation for knowing how to do this kind of stuff. And those are the ones who, who I think won 2025.
Sam Gaddis (04:03)
Yep, makes sense.
Matthew (04:04)
Let's turn over, be a little more cynical here. Who or what was the biggest loser in 2025?
Sam Gaddis (04:09)
This one's easy. It's just the inverse. It's the resistors. It's the do-mers. We run into this archetype less and less, but they're still very present in some of the organizations that we deal with. And you just see them around the people who say AI doesn't write good code and resisted at all costs. I love going to Hacker News because it's still about half and half, but there's, there are always,
great takes on the opposite side. And they essentially amount to, you know, I can't believe you guys still think that this isn't working. I'm a senior coder and I use this all the time. We had in one of our clients, I wouldn't say he was a resistor, but you know, somebody who was a bit skeptical of vibe coding, who is writing all of his code now with AI. And I just, think that the...
the people who are still resisting are gonna have a really hard time catching up. And I think they're actually gonna start losing jobs pretty soon.
Matthew (05:00)
Yeah.
Yeah. There's this trend that you're definitely picking up on here. It's like people who form their opinions in January, 2025, I think are the losers. like, they do not see how far things have come since then and the rate of progress. that's the, that's. All right. Mine, I, I again, went a little bit more macro on this and I almost want to apologize off the bat. I don't, I still love this company.
Sam Gaddis (05:13)
Yeah. If you haven't updated, yeah, that's
Matthew (05:25)
I still want them to like us. I still use it constantly. But my biggest loser 2025 is OpenAI. I think if you think about where they were in January of 2025 as opposed to January of 2026.
they were the dominant player. think if you, anybody who was using AI in January of 2025, the vast majority of them had only used chat GPT. know, it's like there was, there were people on the fringe who had tried other things, but in terms of their daily drivers, everybody's daily driver was chat GPT. And now they are in clearly a three horse race. You know, the market share has dipped a little bit.
I don't think they have the first or second place best model, you know, right now. think those go to Opus 4.5 and Gemini 3 Pro. You know, I think they're carving out a niche in terms of the consumer. I still think they're dominant in the consumer space, but Anthropix kind of has made huge inroads in the enterprise. And Google, you know, the beast has awoken, you know, and I think like they poked the bear and
Sam Gaddis (06:20)
huh.
Matthew (06:23)
I'm not sure what the future is. think ultimately it's good for consumers long-term where we are right now. they aren't the rate they were releasing and running in 2024 to 2025. It was like watching Apple in 2010 or before that. And I don't think they're quite there right now anymore. So still a believer, but on a relative...
Sam Gaddis (06:27)
Yeah.
Matthew (06:50)
standpoint, they're the loser.
Sam Gaddis (06:51)
Yeah, that horse race is phenomenal for us. It feels like every time one of them ships, the two others immediately ship something incredible to try and capture the news cycle. And I'm here for it.
Matthew (07:02)
All right. All right. What's your award, Sam, for the best app, best product or app of 2025?
Sam Gaddis (07:07)
I hate to sound like a broken record, but it's Claude Code for me. That is a tool that I use for almost everything now, not just coding. I use it for strategy. I use it to print documents now. I have a little printer and when I want to print something and I don't want to think about the formatting, I just tell Claude to find my printer and send it there. So it's Claude Code.
But I would also add to that whisper flow, which I think I've talked about here before. And I totally recognize that people probably find it odd to talk to a computer. My fiance has become accustomed to it, but I'm talking to my machine all day long. It's just an input method that is twice as fast as typing. And it's also nice if you're a lazy person like me. It's just, you just kind of sit there and hold a button down and chat away.
and you don't have to be articulate, it'll figure it out. Whisperflow is nice relative to the native Mac dictation because it corrects any errors, it does nice formatting, it'll give you bullet points. So it's great for emails, but it's also great for coding. It also just seems to handle odd words better. It adds proper nouns to its dictionary very quickly if you correct them. So I really like those two. What about you?
Matthew (08:17)
Yeah. I echo obviously all of that. I have a question for you on Whisper Flow and just like your use of the dictation stuff. Do you turn on the voice mode from the AI? Like, so it's talking back to you audibly, or is it just your input mechanism and it's writing back to you in prose?
Sam Gaddis (08:35)
It's just my input mechanism. I didn't even know that it could talk back to you. Do you use that?
Matthew (08:39)
Of course I can. Yeah. You can pick the voice and you can have full conversations with chat GPT or Claude.
Sam Gaddis (08:45)
yeah, yeah, yeah, yeah. I thought you meant from whisper flow. No, I never, no, no, no, I never use voice mode with, with any of the AIs. I have, there's one exception where if I'm trying to use, I have sometimes found utility in the video mode. Have you used this in chat GPT? That can be kind of cool. And it forces you into using the voice mode. If there was a way for me to just use text within that, I would prefer that.
Matthew (08:47)
No, no, no, no, no, no,
Me neither.
Yeah, yeah. Yeah, that's great.
Sam Gaddis (09:11)
But the video mode can be useful when you're trying to troubleshoot some home maintenance issue.
Matthew (09:17)
Yeah, totally. Totally. Yeah. The things I would add to apps of the year, ⁓ granola, which is the thing we've talked about a bunch. It's just the best call transcription service. And I think there's this combo of, know, it just getting things into transcripts is like the hack of 2025. You know, it's like, so either by you talking to your computer, recording your call or whatever, and then using those as your input data for whatever workflow you're about to start.
It's, know, we are literally building our back office around this, this type, this, this pattern, right? This trend of, of transcript to action. ⁓
Sam Gaddis (09:50)
Yeah, we
haven't made a proposal from scratch since we started this business. Every proposal is just take a call transcript and make a proposal. And that's it. It's wild.
Matthew (09:56)
We'll get there.
Yep.
Yep. And then the other one I would just shout out is like, I do think cursor has separated from the competition in terms of IDs, in terms of, you know, ⁓ development, environments there at the beginning of 2025 and through 2025, there were so many low code vibe code tools out there. You know, we've, we've used all of them, lovable replet, a million of them. Right. And I cursor is the only one I still use, you know, they've, they've like, they've come through.
The end, to me. ⁓
Sam Gaddis (10:29)
Are you using
the AI model in that still or are you just using it as an IDE?
Matthew (10:33)
Well, I do both. I often have my Claude code in terminal in cursor. So I don't use iterm separately anymore. And I toggle between that and their agent mode or their complete, and I can change the models and troubleshoot that way. But it's, it's just that it's become the thing that I'm most comfortable in, you know, as opposed to always having to learn a new interface all the time. And, and if you just look at their, they're crushing it in terms of, you know, the revenue and everything like that. So, and even their homegrown model.
you know, composer is pretty good. It's so fast. they, you know, and I think that they've carved out an interesting niche there of just like, we, we aren't going to have the best one because we don't have the same chips as Google, but what if we just have the fastest model for auto complete for easy tasks? And I think that was a really smart move. I thought they made a lot of really smart moves just like that.
All right, next one. And I think this one might be interesting to people. It's best thinker. So who are you reading to stay ahead? And ⁓ who would you recommend, you know, our listeners go to for smart takes, smart assessments in the world we live?
Sam Gaddis (11:33)
Yeah, I'll be a broken record on this one too, but for me, it's always Tyler Cowen on most topics, frankly, but definitely on AI. He's fantastic. I found myself several times this year just Googling Marginal Revolution, the blog, and then AI just to get a synthesis of all of the AI posts that he did there. So think that's great. He also has done more podcasts this year than he's ever done. And I don't know.
half or a quarter of them were on the topic of AI. And so I think there's a ton of useful information there. And the nice thing about the way that he does interviews is I feel like they are much more timeless than all of the other content around AI. It's not as much like kind of what we're doing here, this podcast, just topical. It's much more philosophical and I find that incredibly useful. I also like, I would just say, I would just plug Twitter generally.
Stay away from the for you tab because the algorithm is still horrible and it changes all the time. So it's totally unreliable. But if you can find some good follows and just stay on the follow tab, Carpathy is good, Rune is great. And I would just start there and see who those guys follow and build out a list of AI people. And if you just stay there.
That's where all the news happens. That's where I get 90 % of my content except for marginal revolution. How about you?
Matthew (12:52)
You know,
yeah, in our next newsletter, we should just have a bunch of links to the people we follow on Twitter so that they can, people can go build their own. It's so easy to do, but yeah, I completely agree. I, people bitch about Twitter all the time and I have such a great experience there, but it's because I've curated.
Sam Gaddis (12:58)
Yeah, that's a great idea.
Matthew (13:08)
I only follow a couple hundred people and I add a couple more a year, feel like. Yes, it's my bubble, but it's a very good wholesome bubble. I don't get exposed with all the hate and all the nonsense, because I know what I'm looking at.
Sam Gaddis (13:21)
It's clearly the smartest social media content that's out there. But if you don't have that experience, you're just doing it wrong.
Matthew (13:26)
Yeah. I would shout out Aaron Levy from Box as well. think he's, he's, he's really sharp in this, in this space of stuff.
Sam Gaddis (13:33)
Are you still following
his content? Like, are you still reading everyone? I was super excited about it at the beginning when you first turned me on to him. And now I feel like everyone is this sort of mid long form, but it almost feels written by AI or publicity person at this point. Do you disagree?
Matthew (13:41)
Yeah.
interesting.
I don't know. Maybe I haven't been spending as much time on Twitter the last month or so, which is a good, probably a good thing. But no, I still find, you know, he does a lot of panels. So there's like saturation as well, of course, of just like you're getting the same stuff. I think he just seems like a very sane voice in the room that isn't a huckster, isn't overselling, you know, the capabilities, but, you know, actively trying to transform his business with this stuff. And yeah, I find that.
Sam Gaddis (14:02)
Maybe I just got saturated. Yeah.
Matthew (14:15)
I find that very useful.
Sam Gaddis (14:16)
I think maybe what it is is he's just laser focused on AI and the enterprise. And so the content can feel repetitive to me, but they're probably original thoughts.
Matthew (14:21)
Mm-hmm.
The only other thing I would shout out here is the newsletter TLDR, which is quite good in terms of just roundups of AI news and quick summaries. There's a million newsletters, but that's the one that I find most utility in right now.
Sam Gaddis (14:40)
You know, the other thing that I'll shout out is the Run Point magazine, which you can get on our website. Yeah, so we started to write a book and we were transparent the whole time that we're writing this with AI, writing this with Claude Code. We're using Sonnet 4.5 and then ultimately used Opus. And I kind of realized after I'd completed 17 chapters and 280 pages that
Matthew (14:43)
Great, great, yeah. Talk on that.
Sam Gaddis (15:07)
Nobody would ever read this and I didn't want to read it. I did read it, but I didn't want to. It wasn't enjoyable. And the whole idea that business books have to be that long is just an artifact of the publishing industry. Nobody wants to spend 20 bucks on a book that's 10 pages, even if that's the appropriate amount of pages for the content and for the topic. And so I threw it away and we rebuilt it as a magazine.
And I think this is something that we're gonna try to do quarterly. We'll see how people receive it. But it's illustrated, it's got six articles. You can get it on our website. And I think it's awesome. I really like it. There's, Matthew wrote a couple of the articles. I wrote a couple of the articles. There are big pieces and small pieces. A small piece, for instance, is why executives should use cloud code and how to do that. So yeah, I'm jazzed about this. We just finished it.
And I think it's a pretty great read if you're an executive trying to implement AI right.
Matthew (16:02)
Yep, I'm genuinely proud of it. I hope you are too, Sam. We put a good amount of work into it. think, you know, we're actively trying to fight all things slop here, right? So how do you make something with a sense of craft while still using AI, you know, appropriately to speed up what you're doing and all that kind of stuff. so the visuals are cool. It's a good physical, you know, thing. It's got some heft. It's a great read. I do want to formally apologize.
to you, Sam, and to the people who I inundated. Quick funny story. I was setting up an agent that had access to my email to get people's physical addresses to send this zine to. In my process of setting that agent up, I mistakenly gave it access to our entire address book of Run Point. And the agent sent out an extremely undifferentiated, ugly bad email.
to hundreds and hundreds of people introducing this magazine. People responded thinking that we were selling magazine subscriptions. They had no idea what it was. It was a complete mistake, just like a classic thought I was in testing mode, but the thing was actually live and off it went. So I'm sorry for the people that did respond. Thank you. Your magazine's on the way. And sorry, you Sam.
Sam Gaddis (17:14)
it's all good. It happens.
Matthew (17:15)
All right. Biggest surprise of the year. What was the thing, you know, the morning that you found some new release, some new news story, some new products that surprised you most this year?
Sam Gaddis (17:17)
What's next?
It's not a release, although those did surprise me. mean, the capability of Opus 4.5, even though it's not brand new, still surprises me. Today surprised me. This morning, just working on something. Its ability to one-shot things. But that's not my answer. My answer is the model's ability, generically, 4.5 but also Gemini 3 Pro, the model's ability to produce code that will pass a code review
in a serious enterprise. And that will not have security flaws, that will be vetted by somebody who's not an advocate of AI and can get a score of an A. Like that has happened. And I'm talking entirely vibe coded. mean, vibe coded in the way that we do it, which is thoughtful and, you know, it's not something that you can do without tons of experience, but still vibe coded.
We're interacting with Cloud Code almost entirely to produce every line of code. And it works and it passes code review. That blew my mind. also on existing code bases. No.
Matthew (18:25)
And for the listeners, it didn't work the first time. I think there's an
important lesson here, right? There's lessons along the way and best practices and skills added and architecture decisions, right? That all were required to get, I think, clawed. But I think now you are confident that you can continue to contribute meaningfully to the reproach of this enterprise with a 100 % vibe coded code, which I think is the milestone, right?
Sam Gaddis (18:52)
Yeah, and it's happening probably three times faster than the current process. I think, and that was on the first round where there was a lot more iteration than I think there will be in the future. So I think we'll get up to, I don't know, five, maybe even more times faster. And this is on, I'm talking about an existing legacy enterprise code base, which is just mind blowing to me.
Matthew (19:14)
Yeah. My biggest surprise of the year, I'm going to give to the first Nano Banana Google's image launch. I think the morning that I tried that for the first time, having been somebody who had played with mid journey and early on and, you know, obviously Dolly and opening eyes efforts and image generation, they always felt novel and kind of cool, but really quickly frustrating. And
Never got anywhere close to utility for me or for a lot of other people. feel like they're just sort of party tricks in terms of image generation. feel like Nano Banana was the first time I could see, man, this, this has come along far enough that I see the future now. like I buy into it, that this is going to get good enough and consistent enough to actually produce, you know, long-term benefits for creative people or businesses or whomever. I think a lot of people are scared.
you know, of image generation, because it, you know, takes artisan's works potentially, that it creates deep fakes, all these sorts of things. And it also was like this moment where it seemed clear Google, know, going back, like Google is taking this seriously, you know, that's like that this is, that they're moving ahead and they've only gotten better since then. I think, you know, Sora was another example of, you know, video of something that really impressed me. But I think
Sam Gaddis (20:19)
Mm-hmm.
Matthew (20:29)
2026, we're going to get a lot more actual business application from Image Generation.
Sam Gaddis (20:35)
I've been surprised at how much I've used image generation. When that came out, thought that I wouldn't, I mean, maybe if we were producing like stuff for the magazine, I thought we would use it. But I'm using this a lot now in just my day to day. like, I just started doing woodworking again and I wanted to visualize some cabinets in the bathroom that I might build. And I was talking to Victoria about it. She said,
I want this, I put it in there, and now I have a perfect visual of what this would look like as a finished product. Did I show you the wardrobe thing that I built? So I built something just in a weekend, last weekend or the weekend before, where I took all the clothes in my closet, snapped a quick photo of each article of clothing, took a picture of myself in front of a wall, and then had it style me.
Matthew (21:09)
No.
Sam Gaddis (21:23)
with the clothes from my closet so that I didn't have to think about what to wear. And it pulls in the weather and I specify if I want business casual or whatever. And it gives me studio shots of all of the individual articles of clothing and then me in a studio. I look a lot better than I do in real life.
Matthew (21:40)
Ha
Sam Gaddis (21:40)
but it looks great and it has like genuine, it's useful.
Matthew (21:44)
Did I show you the thing I built? My mom's 70th birthday happened a few months ago and we all went up to the mountains for her birthday, like all the grandkids, everybody was there. And one of the like gimmicky, know, we all made pizza together. We had some planned activities, all sorts of stuff. But one of the things I did was I customized a game of Guess Who? You know, the child game where, know, you all...
Sam Gaddis (22:06)
Yeah, yeah.
Matthew (22:08)
AI generated images of my mom throughout the years of just like pilot Molly and Dr. Molly and baseball player Molly, like a bunch of different ones. And so it was guess, guess Dottie. That's her grandmother's name. And it was just so fun. You know, it took 30 minutes and like, you know, everybody, my daughter thought they were all real. You know, was like, when did, when did Dottie learn to fly planes? You know, all that kind of stuff. Yeah. Applications are both silly and useful, but like the consistency is the thing that wasn't there before.
Sam Gaddis (22:30)
Yeah.
Matthew (22:35)
you know, that like you could, it can generate. Yeah. Yeah.
Sam Gaddis (22:36)
Yeah, you can do it so easily. All
of our little party invites, we're always hosting and just having dinner or friends giving or whatever it is. And now all of our party invites are Victoria chasing me with a knife and I'm a turkey, but my head is on a turkey or something like that. And they look great.
Matthew (22:53)
Ha ha ha!
All right, segue here. Best thing you built this year. We'll do best and worst combo here. Sandwich them.
Sam Gaddis (23:03)
Hmm. Okay, best thing is something that we're building right now.
⁓ so as you know, Matthew, you and I had a conversation a few weeks ago where we said, do we want to try and grow the business in Q1 as much as we can, as fast as we can, or do we want to try and harden the systems that we've got? And we chose the latter. And the reason that we want to do this, we want to be able to scale up our operator engineer practice.
to be able to take on more clients, but do it in a way that is reliable, consistent, uniform, service quality, all of that stuff. And the fundamental underlying tooling that we need for this is essentially CRM on steroids. And so we've embarked on that journey of building that, the dream CRM. At this point, a year into this company, we've already built two, I think, versions of CRM.
Matthew (23:51)
Mm-hmm.
Sam Gaddis (23:52)
and they both worked great. I don't have any complaints about that, either of them. But what we really want is a service business that has the entire back office just on AI enabled rails, where every single call goes into the corresponding account. The account map gets updated so we know what's important to the individuals, what's important to the company.
If the status of a project changes, that is all updated in an automated fashion. If a new project is conceived, we automatically create a project charter with uniform fields so that we can keep track of those. If somebody enters or exits the company, we can onboard them or off board them with consistency that we need. And so this has like been a dream.
project to build since the first time that we had to deal with this 15 years ago. It just wasn't possible. But now we're doing it. We're building it on an open source CRM framework called 20 with the help of our newest operator engineer Ryan, who is fantastic. And I'm really excited about this. think the first thing we did was build the campaign for the
the magazine mailer. So now if you sign up on Stripe, it automatically logs that you are a member of that campaign. It'll then go to the Shippo API, purchase a shipping label for your address, and then print it out on my thermal printer here on my desk. And I just slap it on the box and off it goes. I think the next thing that we'll build is integration into Granola so that we have all of our call transcripts in there.
But this thing is going to be awesome. The amount of AI that we're going to put into this CRM is going to be very cool. So I'm excited.
Matthew (25:30)
Yeah, completely. And I think you alluded to this before, but the, actually the first thing we built in this journey was the proposal creator part of it, Sam. So it was like, right. think that that's my answer here. And then it kind of showed, it showed the model for, think how the rest of the CRM is going to work. So right now, you know, we have, let's call it best practices for how we do discovery calls and things like that. Right. We're trying to get certain pieces of information out of our calls, you know, that we have with clients, just like any organization does. And so.
Sam Gaddis (25:37)
right. Yeah.
Matthew (25:58)
Think about that in terms of the raw materials, proposal creation. We want to know your goals. We want to know your project backlog. We want to know your dates. We want to know all these sorts of stuff. And so as soon as those calls end, we drag and drop, you know, transcripts of the call into our proposal generator. That generator has a bunch of system prompts, so it knows, you know, what it's looking for in those transcripts. And it turns that transcript into a custom bespoke proposal.
that's interactive and thoughtful and takes into account what RoomPoint does in our case studies and all sorts of things like that and generates, you know, at least a V1 of a proposal that is, shocks me every time by how good the first version is, right? And then like, it's pretty easy to edit afterwards, but that was, and shout out to Ryan who built that for us on our team. But I think that's the model that we're going to take to the rest of the workflows within the supercharged CRM of
Think about these meetings we have as the raw input materials and then what is the structured outputs and then how do we actually move things forward. So from standups to retrospectives to charters to any sort of project management stuff, all of that will sort of be baked into our workflow future. Hopefully. I'm very excited about it. Let's talk about the worst thing we built though. You got any, anything you want to?
Sam Gaddis (27:08)
Yeah, it's gotta be everything that was complicated or complex. I did early in the year this big Byzantine rag system built on kind of like a virtual data room. And it technically did everything that it was supposed to do and yet the results were not impressive at all. And that's a pattern that I keep seeing.
What seems to work are simple deterministic workflows with the minimal amount of AI layered on top. And what we were building, especially in the beginning, were complex workflows, especially those that leveraged rag and yeah, didn't love them.
Matthew (27:48)
Yeah, I threw away so much code this year. think that's like, and I think because we came to this from the product side, not the technical side, we have a lot less, what's the right word? We're not as protective over our code, right? Especially because AI is writing most of it. And ⁓ it's just a common behavior we have, which is to try something three completely different ways at beginning of a project and throw away two of them. And so I just went through this process of just cleaning up my local directory.
Sam Gaddis (28:09)
Yeah.
Matthew (28:14)
of just cleaning up my machine of all these stuff. you know, there are so many V3, V4, V5 of the same, of the same projects. And I think if our clients knew, don't often see any of that stuff, right? They just see the, they see the one that ended up working, you know, but, but the, the failed paths to get their art scene. and yeah, I think that's just part of the, part of the process now. It's just trying multiple times until you get somewhere.
All right, that's the end of our 2025 recap. Now we're going to move a little bit into 2026 predictions. So what's the best way to do this? We want to make sure we are thinking about not just saying the most obvious things, the trends will continue, but I want to know from you, Sam, your biggest, your predicted biggest winner for next year. So who is positioned to have the biggest breakout in 2026?
Sam Gaddis (29:00)
I think it's going to be product managers. I think we're starting to see this within our clients. The product managers are watching what we're doing and saying, hey, I can do some of that. And I think that's great. We're encouraging them as much as possible. And this could be this, don't have to have the title of product manager, but it could be a, you know, semi-technical VP or an operations person. But I guess I'm referring to the people who have not coded much before or haven't coded in a long time.
but who are curious enough to roll up their sleeves, learn cloud code, and start trying things. I think those people are gonna be shocked at how much they can accomplish.
Matthew (29:36)
Yeah. I think it's my answer is Google. And I think, you know, just in terms of thinking about the big players here, who do I think is going to pull ahead? We've talked a bit about how they turn the ship around in 2025, and I think they are poised to really crush 2026. We will see if they actually can. Obviously, there's a lot of execution risk in a lot of this, but I was very hopeful by the
Anti-gravity, know, like basically demo that they released. It's free and has no monetization strategy. And I think that you can really only think of it as a beta right now. But I think there is a really interesting product there that if they, they obviously they can definitely screw it up. It's screwed up lots of things in the past, but I think if they play their, they have the best cards and if they play them appropriately, they can win massively.
And I think with Sergey Brin back in the fold, they seem to have more of that, you know, founder energy going and they can make their own chips. They can make the best models. They can integrate into Google docs and workspace. They, I think they are poised to win massive.
Sam Gaddis (30:36)
sense.
Matthew (30:37)
All right, what about biggest loser? Who's going to lose 2026?
Sam Gaddis (30:41)
I think we'll start seeing the downfall of about half of the SaaS market. Certainly the big SaaS entrenched players will stay. They'll be able to ride on their existing agreements, if nothing else, for long time. But.
So many people are already making the choice to build versus buy. And I think we'll see that continue because why on earth would you pay $100,000 for licenses to a SaaS product when you can build it for 80? And you have to pay that $100,000 every year. It just doesn't make any sense. It's such easy math.
I'll add to that actually is making me think of, think, add back into the winner category would be open source frameworks. Suddenly those just increased a lot in value because you can pull them down and modify them to your needs.
Matthew (31:23)
Hmm.
Yeah, yeah.
Totally. Yeah. My loser is, ⁓ low skill home services is the category I'm putting this, but it's your plumbers and electricians and it's all of those jobs that we were told, are actually future-proof because of AI, you know, learn how to do stuff with your hands is what, you know, I think, I actually think that's dead wrong. ⁓ because it's so much easier to be a DIYer now than it's ever been.
because you can point your camera at whatever problem you're having as a homeowner and ask the AI how to fix it. And you can do a lot more than you used to be able to do without actually having an apprenticeship or spending a lot of time on forums or watching all those YouTube videos. I'm not a particularly handy person, but I just went through this process of my furnace breaking. I went down there and through showing pictures to Claude and having conversations with it.
replaced the pump, went to ACE Hardware, knew exactly the part I needed, know, did it all, right? And I, yeah, I was not capable of that. And here's what's even further here is eventually I got to a point where the gas valve needed to be replaced. And the gas valve is the thing that you actually need to be licensed to do. You don't want to mess with your own gas, right? And so I did everything I could up to the point where I needed the professional, right? And so I called the HVAC person to come in, do it.
Sam Gaddis (32:28)
That's incredible.
Sure.
Matthew (32:50)
And I was able to effectively negotiate them, you know, to give me the labor only quote and because the piece is under warranty and I actually have it all here because I was able to do all this. It's just like not get taken advantage of, right? Whereas I feel like if that had happened three years ago, I would have been out thousands of dollars because they would have upsold me on a bunch of things I didn't know anything about, right? And I just think that's the, that's the.
Sam Gaddis (33:02)
Yeah. Yeah.
Matthew (33:12)
Their margins are going to get squished because there's going be more people, more jerks like me who figured out how to do stuff for themselves.
Sam Gaddis (33:18)
think it kind of makes a smart person super intelligent and super capable. I think that's what it's doing right now. Another category that this makes me think of folks who will be losers as lawyers. And I know there's going to be a lot of disagreement about this, but.
Matthew (33:22)
Mmm.
Sam Gaddis (33:32)
I'm using lawyers a lot right now and I'm using them in several ways and not using them in several ways. like doing a commercial real estate transaction where the numbers are large enough to a lawyer matters. Like I'm not going to risk this deal on some, I'm trying to use AI because I think AI is great. It's not worth it. So definitely use a lawyer for that. However, I also updated my will this year.
And for that routine legal kind of stuff, I don't know, would you, I definitely would not use a lawyer for that. It seems so straightforward. Can you imagine using a lawyer for something simple as that?
Matthew (34:13)
No, I think it's the exact same paradigm as what I discussed. It's the low skill stuff, know, the high velocity. There's a million templates out there. It's pretty straightforward. Yeah. And I think you've got people online that'll tell you, those AI generated wills will never stand up in court. like the, it's all documented, you know, your states in idiosyncrasies are all out there. It's like, it's not, it's not beyond the AI's capability of figuring out, you know, how to get there.
Sam Gaddis (34:39)
That's cope. I mean, it's unfortunate because there will be real consequences for the people who have made their living doing this, but that's cope.
Matthew (34:45)
Yeah. All right. This is related to that those last two, but go a little bit farther out in the limb. So that's your biggest winner and loser. But what surprises do you see coming? What what bets do you want to make this year that you you give a lower likelihood to but you think are more likely than other people?
Sam Gaddis (35:04)
Well, I don't know how many of our listeners care about this, but the topic of AGI comes up constantly when people are talking about AI. The idea of artificial general intelligence that can kind of handle anything and people constantly debate about the definition of AGI. And I don't really want to wade into that.
philosophical morass. But I think AGI is basically here with Opus 4.5 or Gemini 3 Pro. I think we have AIs that, like I was saying a second ago, if you are smart enough to take advantage of these tools, and it's like, it's not easy, you know, you have to learn how to ask them questions. There's skill that you develop, but it's also
a fundamental underlying curiosity. You have to have a vocabulary that supports asking the right questions. It's not easy, but you don't have to be super smart. I'm definitely not a super smart person. You just have to be smart enough to cross the threshold where you can ask the right questions. And then this, the tools that we have now with Opus 4.5, whether you use it in the cloud UI or cloud code or whatever, make you superhuman.
No question about it. And that coupling to me is essentially the same thing as AGI. Like I don't know what more to expect from it. It might get better. It might one shot more things than it's one shotting now. Sure. It'll have access to more data and be able to look further back in my history and read my emails and stuff. Great. But to me, we are already essentially at AGI stage.
Matthew (36:39)
Okay. All right. My contrarian prediction for 2026 is that, ⁓ vibe coding, term vibe coding will die. that it's going to get a rebrand and it's going to be the same thing, but all the negative connotations are going to get washed away with the term vibe coding. It's like you've, you've got right now, you've got people use that derisively constantly, right? It's like this was vibe coded, which is code for.
⁓ you, nobody knows actually how it works and it's probably going to break in production and all that sort of stuff. And we've talked, spent this whole conversation talking about how that's not necessarily true, you know, that you can get great results through it. That most people who have, you know, updated their priors and tried stuff recently are enthused by this. The term itself though needs to go away. And it's, you know, before vibe coding, the term was prompt engineering. And it's prompt engineering had all of this like.
connotations of, this is going to be the new skill set of the future. You're going to hire prompt engineers. This is, know, and now all that stuff sort of came to pass, kind of sort of like went away. The actual skills are there. People are better or worse at this than other people are of actually collaborating with AI. And I just think we're going to have some new category of a brand that is going to be a job title of the future. It's certainly not going to be VibeCoder though, but it's going to be up to, it'll be probably McKinsey or Accenture or somebody is going to make some new name.
that's just a refresh of this, but all of sudden that's gonna be the thing where you see departments called this, where you see, you know, and you see degrees that rely on this in colleges. I think we're one brand away from that.
Sam Gaddis (38:10)
Yep,
that makes sense to me, for sure.
Matthew (38:12)
let's wrap this up. Let's wrap this up with a final question, which is your resolution and goals for 2026 for run point. What do you got?
Sam Gaddis (38:12)
It's getting long.
I want to make the best back office AI in the small business category. I want to have our consulting practice just be killer. you know, we, build the coolest stuff for our clients and we have pretty cool tools. But when we're working for clients, obviously we're constrained in what we want to do, what we think they should do for good reason. But
To be able to build the most badass CRM ERP that does everything in the most efficient way with AI is very exciting. And I'm gonna knock that out of the park.
What's yours?
Matthew (38:57)
Mine,
mine is a personal thing, but I'm also going to force it on the whole company. So I hope that's okay with you too, which is to make learning a priority. So, you know, we're often, I think the reason why we got into this is because we're in some, in some way, or form auto didacts. Like, you know, that's the term people who are able to teach themselves things are interested, curious enough to do that. And
I think AI is this incredible source for learning. If you know how to, you know, prompt it in a way that actually increases your knowledge over time on subjects. And we're always grasping a little bit further than our prior capabilities. You know, it's like, you know, we're building stuff with code that I understand more than I used to, but I still don't understand as well as a computer science major, you know, and I wouldn't, I wouldn't claim it right. But there are, I think if we.
If we really focused on not just letting the AI outpace our growth, but we're upleveling our skills along the way with it. think that's, you know, that's the, that's the self-improvement for the entire business and for ourselves. And so like an example of this is I just did this yesterday. I want to, I want to be better at math. just like, you know, it's the thing that I always liked it growing up, but I went liberal arts in college. I never did any sort of advanced math. And so I had.
I had AI build me a curriculum of, if I want to spend a couple hours a week, nothing crazy, but what video should I do? What book should I buy? How do I actually learn this stuff over time? It's incredibly good and thoughtful at packaging those sorts of things of just like, this is the most bang for your buck. Learn these concepts, not these concepts. You know, all that stuff's out there. Watch these videos, sign up for the course, all that kind of stuff. then I think if you think about our
Sam Gaddis (40:32)
That's so cool.
That is so cool.
Matthew (40:36)
Our operator engineers, right? We want more people like this. We want people who can share their knowledge and get better. So if you think about that part of our business as, you know, training paths, learning paths, that kind of stuff, I think that's the, this community could be pretty great at, you know, at sharing what we're good at and how we are all improving, you know, individually. So that's my resolution.
Sam Gaddis (40:57)
Awesome.
Wonderful resolution and the perfect spot to end.
Matthew (41:01)
All right, well, we had a great 2025. I think we're gonna have an even better 2026. Looking forward to it all, Sam. Talk to you soon.
Sam Gaddis (41:07)
Yeah,
likewise. Talk to you later.
Matthew (41:08)
Bye.