AI After Dark

00;00;00;03 - 00;00;22;08
Unknown
When I first asked you this, I, you were you were like, very clear that like, hey, look, Natalie's not an I native company. Yeah. We are introducing some eye elements. What does that mean? So. And this probably kind of covers my whole philosophy on technology, and I think I get strong influence from my time at Microsoft.

00;00;22;08 - 00;00;56;13
Unknown
But, I have always had the kind of fast follower perspective rather than the, bleeding edge first adopter. And there's a couple reasons for that. One, I think it's more pragmatic and conservative. We can talk more about that later. But two, it gives you a better chance to, understand the strengths and weaknesses of, you know, different tools, the environment, what people are doing.

00;00;56;13 - 00;01;28;21
Unknown
It's working, what people are doing that's not working, and also dodge the pitfalls that I think a lot of people fall into when they are being part of the early wave. I also think there's a huge I was my background at Microsoft was product focus. So I thought product first and then implementation second. There's a huge interest, I think, for engineers to grab the shiniest new toy and try to slam that into a product solution, even if the market doesn't want it.

00;01;28;24 - 00;01;51;09
Unknown
And so you can have tons of technologies that I think are great fits and make all the sense in the world, but if the market's not ready for it, it's not ready for it. And so I think for us, for me at utility, when I look at the kind of current AI revolution, because this is only the most recent version of AI.

00;01;51;12 - 00;02;21;00
Unknown
But it's, you know, talking about the lambda genetic workflow, I think the, Interesting thing is there's a lot of places that you can just plug it in and it feels like magic. And because that first time demo, that first usage feels so magic, you immediately start thinking of all these things that it should just work. But once you start trying them, they don't always just work.

00;02;21;00 - 00;02;44;27
Unknown
And so a great example for us, we spent years building individual PDF parsers to parse every different kind of utility bill that a provider might dispatch. And there's all sorts of problems with that. From a technology perspective, if they start putting a full page ad in their PDF, your parser breaks. If they change the wording they use, it breaks.

00;02;44;27 - 00;03;12;26
Unknown
If their first bill looks different than their last bill, it breaks. That's one of the amazing promises of LMS, is that it can read the intent and not the exact, you know, parsed token. And so we looked at that, we're like, great, let's slap that in there. No longer need all these other parsers. Let's just go AI this instead, and it'll just be better.

00;03;12;29 - 00;03;43;13
Unknown
And it was at some things, but it was also way worse at other things. And you start to get on this, treadmill of. Okay, well, is am I using the tool wrong? Is the tool not good enough for my use case? You know, if I just write a better prompt to it, be better next time, and then you realize that you're kind of spending all your time on trying to make a tool work that just might not be the best tool for the job you're trying to do.

00;03;43;16 - 00;04;08;01
Unknown
And so at the end of the day, whether I am using AI or hard coded PDF parsers, my customer does not care. They care that I have accurate information, I have accurate usage info, I correct, you know, data to power their workflows. They don't really care how I get there. Yeah, yeah. And so people. Right. Exactly.

00;04;08;03 - 00;04;42;10
Unknown
And so I think, you know, when I think of, you know, you originally asked the AI first versus AI natives, that spawns all sorts of ideas for me because there's kind of two sides of the adoption there as well. There's like the product focus, which I think a lot of people typically think of. And then there is like the internal adoption and how you drive your internal workflows, which for some things like LMS is going to be massively interesting.

00;04;42;13 - 00;05;14;24
Unknown
But you might not even change your product. It might just be changing how you develop your product. Like our example, my customer will never know if I use AI under the covers or not. All they know is my product is better than my competitors. And so I think the, you know, the answer there is I prefer the fast follower approach.

00;05;14;26 - 00;05;41;05
Unknown
And kind of always have if you look at this wave, I think there's been several times in kind of software development history where there's been a major paradigm shift in how you build products and how you build, your company. The most recent ones you can think of, blockchain huge wave was one of the bigger, bigger busts, some would say, where everyone said that's the future, everything's going to be on blockchain.

00;05;41;05 - 00;05;58;26
Unknown
You know, some people still say, some people still say it. And it was super hot for a while. Everyone tried to slam it into everything. And then all of a sudden you found yourself saying like, okay, if you're hosting your own blockchain, isn't this just a really expensive database? You still have all the control over it. Why are we even doing it?

00;05;58;28 - 00;06;30;18
Unknown
And there's some solutions where you cannot get around it, and that's just a great shit. It's a great tool for the job. But suggesting that you should be a blockchain native company for every type of product in every industry, I think is misguided. Before that, mobile natives, big revolution. You have a lot of companies going from, their product primarily being in the web or online to mobile first, potentially mobile only, in some cases iOS first in iOS only.

00;06;30;20 - 00;06;52;19
Unknown
If you think about all of the social media apps, all the dating apps, they don't have websites. It's all mobile. And there's big pros to that. And there's some cons as well. You cut out a massive part of the market if you are mobile only at the very beginning of that curve.

00;06;52;21 - 00;07;14;10
Unknown
But I think for some companies it makes a lot of sense. It's the right tool for the job. Same with the revolution from, you know, desktop software to web based software, where, you know, you used to ship out CDs and version updates and then all of a sudden it's a browser based thing. You just go to a website and it's updating all the time.

00;07;14;12 - 00;07;38;01
Unknown
And so each of these, I think, has product implications, like what is my user experience? I no longer have to put a CD in my computer. I no longer have to go sit at a computer. I can just pull it out my phone. And then some of them are kind of internal product development or internal operations impacts is a is that kind of the lens that you look at utility through.

00;07;38;01 - 00;07;57;29
Unknown
Right. It's you know, to your point, it doesn't have to be AI. It doesn't need some context. And there are there are situations within the business where having contextualization, being generative in some sense and then eventually genetic, makes a lot of sense. But for right now it's it's just like modular finding, identifying the pieces that make sense.

00;07;57;29 - 00;08;32;22
Unknown
Because the way I think of it is for us are customers don't particularly care about that. From a product perspective, our product is solving a problem for them, not so that they can interact with it just so they don't have to think about it at all. And so rather than having 10,000 bills come through your accounting team and then paying them all and doing the accounting for them, we pay the bills and then we send you one thing that you can then let your accounting team go off and do their process for.

00;08;32;25 - 00;08;57;19
Unknown
And so there is a light opportunity, I think, for us to build some AI tools that will help teams explore their massive mountains of data better. But primarily no one actually wants to go looking at their utility data. No one wants to go look at utility bills. The only time they care is if there's a problem. And so at that point, they're looking for a very specific thing.

00;08;57;22 - 00;09;26;19
Unknown
And so our core product isn't necessarily giving someone a AI experience or a better interface, which I do think a lot of modern AI tools offer the kind of natural language chat based interface is better for a lot of things. But in our case, the perfect user experience is they never even log in to our website.

00;09;26;22 - 00;09;48;05
Unknown
So then for us that means like, what are the backend implications? How do I deliver that experience to you where you never have to log in? Everything just works all your bills are paid on time and you never have any utility problems. That's where I think we have tons of opportunities to use AI, and there's also tons of places where any kind of failure is unacceptable.

00;09;48;07 - 00;10;07;12
Unknown
And so one of the biggest hurdles we've run into with AI solutions is they are not perfect. They're not magic. They have an accuracy rate, they have a failure rate. And so if I told you to go pay 10,000 bills and there's going to be a 92% success rate, you're probably not going to be happy with that.

00;10;07;14 - 00;10;35;19
Unknown
And so we have to build processes and tools that move that number as close to 100% as possible. And the AI native approach would say scrap the humans, scrap the code, scrap the hardcoded parsers. And just like AI, everything. And if we could get the accuracy rate high enough, we would, but it's not good enough yet, or we haven't figured out how to crack that code yet.

00;10;35;21 - 00;10;58;13
Unknown
And so I think the way that we're approaching it is we have to have the lowest common denominator work at all times, which for us is a human picking up the phone, calling the provider and sorting out whatever the issue is over the phone. Someday that will potentially be solvable. But today there's lots of people who will claim that you can stream real time voice.

00;10;58;15 - 00;11;24;26
Unknown
We haven't been able to make it work at scale. Yeah, especially because, there's a lot of threads as well, actually. I mean, one, you keep talking about being a fast adopter versus being on the bleeding edge. How are you staying on top of specific use cases that you're trying to resolve, and what tools are out there or other, you know, what solutions that you've seen are out there, that are worth trying out, implementing.

00;11;24;29 - 00;11;50;23
Unknown
What are the case? Maybe. Yeah. So the nice part about the broader open source developer community is they love to share what they're doing. You have podcasts, you have YouTube, Reddit, Hacker News, Twitter. There's people all over the place talking about the crazy stuff they're doing. And it's really compelling. You see, some of the things people accomplish and you're like, whoa, if I apply this to my business, we would explode.

00;11;50;25 - 00;12;30;18
Unknown
I think the challenge is, in practice, the demo is almost always better than the actual production application. And what I have found at least is, AI in its current iteration is very successful at Greenfield, not as successful with existing stuff and adapting to what you already have. And so places where we've had. Tons of success is rapid iteration, prototyping, trying something new quickly and not knowing if you want to discard it or not.

00;12;30;21 - 00;12;49;26
Unknown
Places where we've not had a lot of luck is, you know, take our entire billing engine, which is tens of thousands of lines of code, too much to put all in one context window. And fix this bug, or find this feature and change it, or add this rule to improve the configurability. Those kinds of things.

00;12;49;26 - 00;13;13;29
Unknown
It adds more damage than value when not closely controlled by an engineer. Do you find that, like having that many lines of code adds a layer of tech debt that you all you're looking to clean up? And the reason I ask is, I feel like now with all these tools, it feels like people are, you know, putting aside tech debt because they're like, well, I'll just I'll just spin up something else and spin to something else.

00;13;13;29 - 00;13;35;08
Unknown
Yeah. And maybe they're all toys and they don't actually, to your point, productize in a real way, which is really scary to me, I saw I see some claims that I would love to go talk to, like I saw something from a Google engineer who said, like, we spent an entire year building this and then had, oh yeah, tech tool build it into the whole thing in two weeks.

00;13;35;10 - 00;13;49;17
Unknown
And it was like, I would love to know how you ran that experiment. Did you did you read the article after know so so that was the whole thing. It was like in the title it said that. Yeah. And then in the body of the article, it was like, oh, we spun up something that worked once, you know.

00;13;49;21 - 00;14;26;27
Unknown
Yeah. And so it was like, wow, here's the world of the possible. But so everyone sees the the clickbait. Yeah. Which is like we just recreated a whole team of Google engineers in two weeks. But the reality is like if if that becomes all you have, and this is the extreme, this comes back to your AI native question, because I've now been asked by multiple people, like, when are you dropping your whole engineering team and replacing them all with agent like, you know, programmers, when are you replacing all of customer service with AI customer service?

00;14;26;27 - 00;14;46;21
Unknown
When are you replacing? And I think in order to get there as fast as possible, you have to have the AI native mindset, because you have to be pushing the boundaries of what's possible at all times, like the Google experiment. They probably did it in a conservative way because they have a whole team backing it up in case it fails.

00;14;46;24 - 00;15;16;16
Unknown
But if you're a startup, you don't have the luxury of here's my whole engineering team building the product. And then I'm also going to have a side by side product that's more AI first. And if that fails, we'll just fall back to the one that that works. And so I get the appeal. I think there's places, I think there's products where the competition is so high that you should make your play and you should send it.

00;15;16;16 - 00;15;39;20
Unknown
And if that, if you can crack the code, that could be a massive differentiator. But again, for me, it's it's it's not as conservative. You can't guarantee success in my opinion. And so oftentimes when we talk about experiments, we want to run in utility. I'll have other executives on my team say, well, can't we just do an AI solution?

00;15;39;20 - 00;16;07;13
Unknown
Won't that just be better? I've seen all these things on Twitter. And it's like, well, we can try it and it might be better, but it also might be worse. And that means we might put a bunch of time into something that is worth nothing. Or we can make the incremental progress. Continue. Always be experimenting. And I think my rule of thumb there is everything reinvents itself every six months.

00;16;07;16 - 00;16;30;10
Unknown
And so if you haven't touched a certain type of technology inside of six months, it's been too long. So I recently picked up, cloud Code again, because there is a bunch of buzz around how it's since opus 4 or 5. It's like absolutely crazy. And last time I played with it was probably in the summer, and it is massively different from the last time I looked at it.

00;16;30;12 - 00;16;53;14
Unknown
And it is way better. What's sorry, be more prescriptive about what makes it better, because I think, like, it's funny, you know, we're going to venture capital fund. Not a single one of us was using cloud code, even though it was, like, so amazing and all this stuff. And now all of a sudden, I don't know if it's because we've had dozens of proof points that have been hammered into us or last month, but it's like, shit, we got to do this.

00;16;53;14 - 00;17;15;12
Unknown
We got to do the co-work, we got to do all of it and blah, blah. And it's been great. Like, I, you know, I'll do everything from a snack, sign up sheet for my daughter's basketball team to a personalized CRM. You know, that that only I use. And they're still it's so buggy, right? Totally. But in mechanics all the time, you get those eyes light up moments that that emotionally drive the use of the actual platform.

00;17;15;15 - 00;17;40;14
Unknown
But I've never sat down and said, this is what really makes it better. Or yeah, and it's hard to point to any single thing because there's a lot of things that have to come together to make the experience good. One is the underlying model. I think if you pay for the expensive plan and you can use opus 4 or 5 to answer most your questions, it is better at the general reasoning than prior models, and as a result, the iteration happens faster.

00;17;40;20 - 00;18;08;10
Unknown
And so like as an example, you would say, you know, I would like to, build a CRM to manage this kind of data. Ask me any questions you think are relevant, the questions it's going to ask you are probably just better, because the model is better at understanding your intent, what you actually want. And so you get to where you want to go faster doesn't necessarily make the end result better, but it makes the iteration tighter and faster.

00;18;08;13 - 00;18;41;02
Unknown
So that's one thing. I think the interface has improved substantially. And by interface I mean how for me, the biggest frustration I have with these tools is managing context. Because at its core, LM is only as smart as the context. You get it. And every new session it gets wiped clean. They're adding new features like memory is a thing that didn't really exist a year ago, where they will summarize prior conversations you've had and then inject those into your context for you.

00;18;41;05 - 00;19;15;11
Unknown
And so that's cool. If I have a conversation today, it might remember something I talked about yesterday. It also might not. But previously you had to literally grab and give it anything you wanted it to reference. And so some of those kind of built ins are nice. But I think the big thing is the interface and specifically for client code and a developer workflow, it's how it determines when it needs more context versus just use what I have to make the best inference possible.

00;19;15;13 - 00;19;34;04
Unknown
And so, you know, if you're working on a code base and this is why I say it's better at new projects than existing projects, we have a huge repository full of code, and we're not even that big of a company. So even bigger companies have even more code. You can't simply say, go read all my code and now make a smart decision.

00;19;34;06 - 00;19;51;03
Unknown
It'll blow the context window. There's nothing you do about it. You can do some step by step things like read this file, summarize it, and then store that summary and use it for future reference. But now even that is probably too much over the course of hundreds and thousands of files. So you can do that on a project basis.

00;19;51;03 - 00;20;18;21
Unknown
And so you have layers of kind of artifacts that can help manage the context. But it is better at pulling in things when it needs it. As well as if you do blow the context window, compressing that and rolling it rather than just terminating your session. And so it takes a experience that, in my opinion, was previously I spent so much time telling it, you know, I want to do this.

00;20;18;21 - 00;20;36;25
Unknown
Here are all the files that you need to make your choices. Now it goes and finds them, and it's more likely to find the right thing. And it still will make mistakes all the time. And so, you know, you still have to check what it does. But, but I think that's the big difference in the last 6 to 12 months.

00;20;36;25 - 00;21;10;19
Unknown
Right. And they've also added, you know, cloud code specifically has the infrastructure behind skills, where it has built a pattern that is more human friendly to better human interface for giving and managing context for the tool. And so to me, that's all it is, is previously, if you were super diehard and you put a lot of effort into it back in January of last year, you could have built your own tools and tons of people were doing it where you had things that helped you manage context.

00;21;10;22 - 00;21;30;07
Unknown
You know, you have one limb that decides with the context is and then it feeds it to another one that actually does the work. And it's in the to a third check. It's work. And you could build these really complex workflows. But that stuff, as it becomes successful, as people figure out what makes it good, it gets productized, it gets put into cloud code, it gets put into copilot.

00;21;30;13 - 00;21;54;13
Unknown
I'm sure there's a thousand other startups that are built on top of these core technologies that add layers of just better human interfaces to let us interact with the tool. Yeah, because I feel like what I'm hearing right now and honestly how I feel as well is like, you've got the every Lem has its use case that that it tends to perform better at than others.

00;21;54;15 - 00;22;25;07
Unknown
I mean, I feel like ChatGPT is sucks at everything right now, but but but for a while it was, like very creative. And Gemini had a lot of like, you know, had data pipelines that were a little bit more enriched giving YouTube and other things and in obviously for, for building it. Do you find it maybe it's different as a CTO, but do you find that like you use the LMS differently per the use case or, or they use all of them knowing that like, hey, those that product ization is going to help one be a little bit better today, another be a little bit better tomorrow.

00;22;25;07 - 00;23;02;17
Unknown
And to some extent, I think this kind of goes back, though, to the way I approach technologies like I like blockchain, like whatever the new shiny thing is, I'm less concerned at any given time with the product itself, and I'm more concerned with the kind of meta implication on the workflow, or the business or processes. And what I mean by that is I don't actually care in the future if I'm using cloud code or Copilot, frankly, it will probably not be any of those.

00;23;02;17 - 00;23;23;05
Unknown
It'll probably be something completely different. Someone will have a much better way of interfacing with it. And I always want to be in a place where I can use whatever is the best tool at any given time. Which is one of the reasons why, you know, I hate vendor lock in. But it's one of those things where, you know, you want to keep an eye on.

00;23;23;07 - 00;23;47;28
Unknown
What is the difference between the latest and greatest interface and the core underlying technology and what it needs to be successful. And so when I look at these tools right now, if I'm using cloud code, I had a dot cloud MD where you put all of your context for cloud, and if you're using copilot, you have a doc copilot, something where you put all your context for copilot.

00;23;48;04 - 00;24;11;26
Unknown
If you're using another tool, you know, ChatGPT it has nothing. So you have to upload files to it, paste it in directly, give it the context. Regardless, you are giving it context in the form of written text every single time. And so the important thing to me is not so much which of these tools is best at what.

00;24;11;28 - 00;24;36;13
Unknown
And if I'm doing something specific, I will reach for in that moment what I think will be better. But on the macro scale, I care more about how is this pattern going to exist now and in the future, and what are the building blocks that, regardless of the tool, will be valuable no matter what? It's the last thing I want to do is go all in on cloud code, only for them to completely implode.

00;24;36;15 - 00;25;04;04
Unknown
And now I bet the farm on the wrong thing and I lose all my progress. I would much rather say, you know, why do all these things have in common? What is the core underlying paradigm shift that's happening? And how can we position ourselves that we can play with the things, make the experiments, try out the tools, but be building in a direction that aligns with the entire paradigm shift at its core.

00;25;04;06 - 00;25;26;07
Unknown
And so right now, the way I see it is, there's two sides of it. One is the product experience. And like I said, I don't think that applies to us as much right now in the iteration of utility that we have. And so I'm less concerned with how I can take the AI experience that people are affecting and deliver that to my customers.

00;25;26;07 - 00;25;48;13
Unknown
I'm more concerned with how can I transform my business using these tools? And so for us, that means, you know, I don't need to go out and buy a cloud code subscription for every single person of the company. I don't need to go get a copilot license for every one of the company. Does mean that I want people to be trying any and all of those tools.

00;25;48;13 - 00;26;21;15
Unknown
If they want to go use it, great use it will reimburse it. Just play with it and let us know what you think. But what we do need to do is be building written artifacts that can be fed into whatever the future iteration of that tool is. And so if it's taking our code base and building those summary markdown files at different places so that cloud code can better understand our existing code, I want to do that because what I'm positive of is it works for cloud code.

00;26;21;15 - 00;27;02;27
Unknown
All the others are going to do it, and even if it doesn't work in that exact format, it is going to be cheap in the future to say, go take all of my cloud configs and convert them to copilot configs or whatever future config it needs to be. What it won't be able to do is encode the actual business knowledge, for the business context that you need and so a great example of that is when we're talking about utilities, I might say, you know, how many active units do we have active units in our context means you have a utility account and an active status that can produce bills.

00;27;02;29 - 00;27;25;15
Unknown
And if you say that to an Lem right now, give me active units, it's going to look for active is true or some kind of flag or some. It has no idea what that business context is. But if you can start taking that knowledge and encoding it somewhere that is accessible, then that is building compounding value towards today and future tools.

00;27;25;18 - 00;27;52;01
Unknown
Another great examples MXGp servers all the rage. For probably the last year everyone's been building them furiously and I love the idea. The underlying idea is so good. How do we get this land interface to talk to third party tools? It would be great if it could query my database. It'd be great if you could go visit this website and book me a reservation.

00;27;52;04 - 00;28;26;04
Unknown
Is MCP the solution that will still exists two years, five years, ten years from now? Maybe. Probably not. I think anthropic already posted. Document or a blog or something talking about how MCP was kind of a mistake in that it is a good starting point, but ultimately it's bloated. One of the big problems is if you have a bunch of MCP servers, you're going to load all of them into context, and it's hard to manage what's in and what's out.

00;28;26;04 - 00;28;50;05
Unknown
And so you might blow your whole context window on just figuring out what tools your agent has access to. And so I think you know, the idea is pure the idea of how do we make our data accessible by an LM is a good idea. Does it have to be MCP? Maybe. Maybe not this version in this slice of time, that's what it is.

00;28;50;07 - 00;29;10;12
Unknown
But in the future, that could change. I mean, I think they were already saying that, like, really? Where what Claude needs to be great is to be able to write code to perform repeatable actions. It's like, oh, so you want an API, you want SDK, an API. So Claude can write code, keep track of scripts, and that kind of yields the whole skills workflow.

00;29;10;14 - 00;29;34;02
Unknown
If you're familiar, where you can literally have a reusable script that you teach Clyde that like, hey, when you want to go access, you know, bank transactions from Mercury, call this, this is going to hit the Mercury API. Don't bother with the Mercury MCP server because it takes way more tokens. It's slower, the traffic's worse. You're bogging down the model with things it doesn't actually care about.

00;29;34;05 - 00;29;56;22
Unknown
You're trying to do a transactional thing. Go get need this info. And repeatable. Right? I mean, it's the same process every single time. And so and so those kinds of things, you know, we're kind of going full circle on. But but the point is that like, you know, if you overindex on, we could have said AI is the future of everything.

00;29;56;28 - 00;30;17;24
Unknown
We need to build the utility MCP server so third parties can call into utility for get building out our API. It's the MCP server. Every company is going to want to use utility MCP. They're not going to want to use our API. We could have done that and we explored it. And we do have some experiments going where we're playing with that.

00;30;17;26 - 00;30;43;28
Unknown
But the more important thing is how do we make our stuff available to the third party? And so that's kind of the way I've been looking at it is it's not necessarily which product is the best because it's going to be different six months from now. It's more of what concepts are more than likely to endure, regardless of the production.

00;30;43;28 - 00;31;14;18
Unknown
And so when you say make our stuff available to to third party, are you looking at it from the lens of distribution? And I guess what I'm hearing is, hey, yeah, we're continue to experiment and build and we have our processes for, for, you know, sandbox environments or trying this in that trying, you know, whatever. But really the use case for a lot of AI tools right now is like, how do I or, or at least thinking through a first principles side of things like I need distribution, it can't just be if you build it, they will come.

00;31;14;20 - 00;31;45;24
Unknown
How do I how do I leverage there's there's a bigger opportunity to leverage tools for that than there is for actually building product. Yeah, I think I think kind of more what I was alluding to is because our systems are, heavy on the op side internally, like, how do we make our all of our data and all of our systems and all our business logic accessible through whatever interface it needs to be?

00;31;46;01 - 00;32;15;24
Unknown
Right now, we build all of our own web apps, and our internal apps team uses them to upload bills, process builds, manage utility accounts. And so if you want to say, create 100 utility activations for a customer, you go to a very specific page that we have built for that, and you find all of the houses you want, you check all the boxes you pick electricity, water, gas, whatever, and you hit submit and then it goes off and runs database queries to create the stuff you can imagine.

00;32;15;24 - 00;32;48;01
Unknown
Instead saying, I'm in my cloud code window or ChatGPT or whatever it is. I want to create 100 utility activations for these houses in these types, and then it is able to communicate with us. We have made those capabilities accessible to outside of our control application. Again, part of the problem there is like the the Macfie MCP interface doesn't do great with permissions and you can build around that.

00;32;48;01 - 00;33;11;06
Unknown
You can subsidize that. But the more of that you build yourself now and hack around the constraints of the existing products that exist. Once those products move forward, you're now going to strip everything out and redo it, or live with your custom solution forever. And so, you know, there's trade offs there, like being on the bleeding edge.

00;33;11;08 - 00;33;51;17
Unknown
There's some definitely upside, but there's also an upside to if it's not the most important thing, do the experiment, be familiar with it, know the capabilities, but wait until it's ready. Yeah, yeah. I just feel like I feel like we're, like, barreling towards, an environment where everything becomes the cursor of X, where it's like your, your, your customer base ends up engaging either because they've been educated through just some more generic B2C type, you know, Lem engaging in a chat based interface with your product, especially if that's the interface that you're using and saying, hey, I want to do X, or I want to do Y.

00;33;51;19 - 00;34;22;17
Unknown
And it's it's like creating its own application on the back end to be able to execute on that, that, you know, I mean, in an ideal world, okay, you can take back and productize or build into the platform and all that kind of stuff. But we understand the, the difficulties of doing that. Yeah. Do you see that being the direction you're going, particularly you working with like presumably not hi tech property managers that are just like they're like, man, I've got bigger fish to fry than like having them worry about building, you know, a tech or engaging with the tech platform.

00;34;22;22 - 00;34;47;01
Unknown
Yeah. So this is an interesting question because, you know, part of me is looking at how this impacts utility right now. Part of me is also just thinking about all the time, how does this impact everybody? Like my wife's VP of finance, she has had now multiple pitches from startups that are trying to convince her that, you know, whatever you're doing here is our purpose built AI solution for you.

00;34;47;04 - 00;35;20;21
Unknown
She just sat through one that was going to help them close their monthly books, and it was like, this will be an automated agent. It will close your books for you every month. You no longer need your, you know, staff accountant to do this process for you. You'll save all sorts of money on labor costs. But it's a great example of like, I just don't think that type of user is ready yet, and I don't think the interface is good enough yet, because that type of non-technical user is not interested in figuring out how to teach it.

00;35;20;24 - 00;35;45;22
Unknown
Yeah, not interested in building skills. It's not interested like it's a completely different type of user. And so in order to serve them, I think there will be tons of companies that try to go build purpose built point solutions that are this is your interface for this type of your sub agent, if you will. And that will be great for the early adopters.

00;35;45;24 - 00;36;19;20
Unknown
But I think in the long term, all of that will get consumed by whoever is able to figure out how to build the interface that dynamically pushes and pulls on whatever it needs to get the specific job done that's in front of the user. And that might happen soon, and that might not happen soon. We might have some real technical restrictions that stop us from getting somewhere like that in a reasonable amount of time or compute.

00;36;19;22 - 00;36;41;05
Unknown
And so that's going to be, I think, for me, the interesting thing to see how quickly does that side develop? Because at its core, if you want to use cloud code to automate something, you still have to know a lot about how it works, what its quirks are, how to avoid the common pitfalls. My wife and I are going to do that.

00;36;41;07 - 00;37;09;18
Unknown
The average business user is not going to do that. My customer service team is not going to do that. I think there's a huge advantage for the startups that can put together teams that are AI native, where every employee thinks in that new paradigm shift of not how do I do this task? But how do I build artifacts that allow an agent or a model or whatever to help me do this faster in the future?

00;37;09;20 - 00;37;33;07
Unknown
I love that from an organizational design standpoint of like how I think about AI, native is every if every employee, every person I hire is building artifacts versus just executing a job. Yeah. And it will work for some of them. And I think those companies will see crazy results, assuming that they still have all the other things that you required to have to be successful.

00;37;33;07 - 00;37;52;22
Unknown
You still have to have a good product. You still have to have. Yeah, good economics. But but like in terms of business development velocity, like if your sales team can do that, they will be better. If your marketing team can do that, they will be better. If your ops team can do it will be better. Accounting team. They will be better.

00;37;52;22 - 00;38;21;10
Unknown
And so I think that is an interest. And we were talking about today how, you know, to what extent does our hiring process change to expect some of these skills? Not that I expect you to be a, you know, three year expert in using cloud code, but I expect you to solve problems in a way that takes the repeatable parts and offloads them as part of your solution.

00;38;21;15 - 00;39;07;08
Unknown
It's not good enough to give me the algorithm I want to to right. You have to also develop artifacts along the way. And if you're not doing that, you're thinking about the problem. Wrong. Yeah. You're ignoring the meta problem, which is that continuous progress. Yeah. Yeah, I, I, I mean, I think, I think what, what I think honestly, the thought that kept coming through my head was, was like, when you think about the step change that has to happen from an efficiency standpoint for someone to go through the disruption of, of, like, ripping out something they're doing and, and replacing it with something else, it's yes, the, the tool itself has

00;39;07;08 - 00;39;34;16
Unknown
to be able to execute in a much more efficient fashion, but it also has to remove a bunch of friction points. Right. And I think we definitely are at a point right now where it feels like, okay, we can build a lot of really neat stuff, internally that makes things faster, that makes things, you know, you're able to execute on a, on a higher volume, but we're not thinking as much about how do we reduce friction points from the end user side, particularly in legacy industries?

00;39;34;18 - 00;39;59;07
Unknown
Like in real estate or powering utilities, or oil and gas and, and, and I, I wonder who's like, whose job does that become in a world where every person is building their own artifacts for their workflows? And then I think the challenge is how do you share them? How do you iterate on them? How do you accept the ones that are good, dispose the ones that are bad, like on an organizational scale?

00;39;59;07 - 00;40;22;11
Unknown
We were talking about this on our dev team last week. It was like, okay, I have my own cloud skill for helping me write unit tests. You have your own skill for helping you write unit tests. You probably should both be using the same one. We should probably move, merge them, use the best parts. What? So why like if the if the if you accomplish the same goal with different tools.

00;40;22;13 - 00;40;44;19
Unknown
Because not every engineer has one God. And so you know, if every engineer on my team is writing unit tests and they should be, but only two of us have built a skill that specifically encodes all of our business knowledge. And again, there's things that are generic that does not matter, like good hygiene around test driven development.

00;40;44;21 - 00;41;08;06
Unknown
Cloud can do that by itself. When it doesn't have context on is part of writing tests is you got to generate seed data. Yeah. Test on something. And for us that has business logic in it. And so if I say write me a unit test that has an active house, but you don't own an active house is you're going to see the wrong data and you just generated a unit test that is wrong.

00;41;08;08 - 00;41;31;22
Unknown
And that's now negative value. And so the more business knowledge you can encode, the better it will be at accomplishing your task. Another one is how we handle authentication specifically or how we pass session tokens around. If you can communicate that to the agent before it goes and does the task, it will be way more successful every single time.

00;41;31;29 - 00;41;47;20
Unknown
And these are things that you figure out by playing with the tool, finding its weaknesses. You say, go write me some unit tests and it gives you a bunch of crap, and then you have to go fix it all. You're like, why did I just bother with this? This was dumb. But if you could have known these things, it might have been better.

00;41;47;20 - 00;42;08;28
Unknown
So you try it again the next time, and it's better the next time. And then eventually you end up in this place where you're developing these assets, where in cloud code land, it's skills. In Copilot, it's just markdown files. But in all of these, the valuable thing is what is the asset you're generating to make the agent be better?

00;42;08;28 - 00;42;31;14
Unknown
The best analogy I've heard so far is treat, these a genetic AIS as high school interns. Like they show up, they're super excited. They'll work 24 over seven. They're amped. The core, but they don't know everything about your business. And frankly, they don't know anything about your business until you teach them. And unless you teach them, they're not going to do it.

00;42;31;17 - 00;42;41;07
Unknown
And they're going to push the work and they're going to be confident about it. They'll say, hey, I did it. It's done. You're like, great. And you read it. You're like, this is garbage.

00;42;41;10 - 00;43;05;12
Unknown
So the goal is how what would it take? What were the core difference between a high school intern and an agent is it can consume data at an incredible rate. And so what kind of information would I need to put in text to make that high school intern, assuming they read it all, consumed it all, followed it all, elevate themselves?

00;43;05;15 - 00;43;24;27
Unknown
And the problem is, okay, if you gave it everything you possibly could do in the entire world's data, it would blow the context window. So you can't do that. And so you're like, what am I about to ask the intern to do? And like The matrix and you to fly a helicopter, plug in the helicopter skill, upload that context, go fly your helicopter.

00;43;24;27 - 00;43;49;24
Unknown
And when you land, eject that and do your next task. Yeah. And only focus on that. But that requires you to have all of these standalone artifacts that are well-written or accurate, because if those are wrong, it will destroy everyone's workflow. Another great example, yesterday, was trying to help one of our PMS more quickly. Check out PR so they can test whatever the code is.

00;43;49;24 - 00;44;17;27
Unknown
Look at it, provide feedback, and move on. And so put together this little skill that made it easier for, cloud code to do all of the operations to grab the branch, pull it down, build the code, whatever. Run it, launch the server, launch the web app, all the things. And because when I was building it, it used my local machines, some hard coded parts on my local machine to build the skill.

00;44;17;28 - 00;44;38;08
Unknown
When I transferred that to him, he tried to use it. It didn't work because it broke using things from my laptop. And, you know, it kind of figured it out, but it was wrong. It was just wrong in the skill. And now it's like teaching the high school intern the wrong thing. Everything he does is going to be wrong.

00;44;38;10 - 00;45;08;01
Unknown
And so, so I think the kind of the, the meta level for us, for me right now is how do you build high quality artifacts, how do you collaborate on high quality artifacts? How do you share those with other people in your organization or outside your organization? Like if I wanted to send you a bunch of skills in cloud code that I had built today.

00;45;08;03 - 00;45;29;26
Unknown
Dropbox. Like a USB drive, right? Like you're you're transferring files. If I update those, you don't get those updates. You could have a git repo, you could make packages. There's all sorts of things you could do. But if you look at it that hasn't been productized yet, you can see where it's going, though. Eventually the interface will support that.

00;45;29;29 - 00;45;54;01
Unknown
You will say, you know, I want the unit test skill, and I might get the one that's from anthropic or the one that's made by Bowden, or I might, you know, subscribe to this one and follow it. And kind of like extensions, or add ons that you have any other software and they auto update in the background and you can pin your version if you want to, and all of those things, you could fork it, make it your own.

00;45;54;06 - 00;46;17;24
Unknown
All that stuff will exist someday. It just doesn't exist yet. And as a result, it's tedious to use at scale, especially in an organization. If I wanted to say tomorrow all 100 people in utility, you're all using AI for everything. We're going to be AI native. There are some logistical hurdles that have to be solved just to make that work.

00;46;17;27 - 00;46;38;27
Unknown
And all the time you spend doing that, you're configuring tools. You're not actually making progress. And so this kind of comes back to my fast follower stance of I'm going to play with the tools, I'm going to use them, I'm definitely going to use them where they're useful. Am I going to force the rollout to every person in my company and become truly AI native?

00;46;39;00 - 00;47;01;02
Unknown
Probably not, because I think in the next three months or six months someone's going to solve that problem for me and so I can spend more time focusing on our business problems, which no one else can do. And then the more generic solution will come and we'll adopt it when it's when it's ready. Yeah, yeah. And maybe this is, I guess, fly in the face of what you just what you just said.

00;47;01;02 - 00;47;21;29
Unknown
But I like the high school in turn. Analog where. Okay. Yeah. You teach the high school intern everything you know about this industry, about your business or whatever. But what's the issue with every young person relative to someone that's had an experience, is that experience comes with intuition, comes with taste, comes with the ability to, like, recognize things that don't require like an inherent context.

00;47;21;29 - 00;47;43;08
Unknown
Just because you've been going through these motions or seeing so many different edge cases that allow for you to to have that intuition. How in a world where you're building, you know, workflows that you say, okay, I want this to be the workflow that we use across the business, or maybe even a workflow that that is focused on the actual business problem.

00;47;43;10 - 00;48;01;26
Unknown
How do you get it to iterate on itself in a way that you can take that high school intern and make them a college, you know, or even someone that's got the intuition of someone that's been in the industry for ten years. Yeah, I love this idea because this gets even more meta, right? Like, how do you make the agent train itself?

00;48;01;26 - 00;48;31;07
Unknown
How do you make it recognize when it's wrong or weak or, you know, spend time learning and then taking those learnings and fueling make in, I don't know. I don't know how you do that yet. I think there's lots of good ideas out there that are formulating, like, you see, companies that are plugging in data that standard agents don't have access to, and they're indexing it and they're making it faster to access, and it makes context more manageable.

00;48;31;10 - 00;49;05;27
Unknown
I think some combination of that. And then these multi-agent groups where I've seen some examples of people using the the committee or the council where it's not one agent doing a task. You spin up five agents, each of which has a different kind of context or slant, and then by committee they improve themselves. And then of course, after each iteration you need some output artifact such that you don't waste the time relearning the same things over and over again.

00;49;05;29 - 00;49;40;19
Unknown
But I think ultimately, at its core and this is, this is the extent of, of my understanding of, of the technology. But at its core, you're processing a lot of text and predicting what comes next. And in order to make that better learning, like actual learning, it's not like reinforcement learning in other machine learning approaches where you can literally make the model change.

00;49;40;21 - 00;50;01;20
Unknown
The model doesn't change from iteration iteration only its context changes. And so the question is, how can you continue iterating on that context for each given task type to get to a point where the generic magic solution is, I ask you to do something as if you were an employee, not a high school intern, not a college intern.

00;50;01;20 - 00;50;28;21
Unknown
You're just an utility employee and you know what context you need. You know when you need to go ask accounting. You know when you need to go ask legal. And you've learned that from experience. And so you can imagine a world where if you have enough checks and counter checks and self-improvement loops going, that these things could continuously improve the context they use to answer questions.

00;50;28;24 - 00;50;48;26
Unknown
But I think to get there, you have to set up these kind of recursive workflows and who knows where those end up. I think someone will crack it and it will be incredibly powerful. And I think plenty of people will try and they'll end up in a place where you're not getting much better than what you have today.

00;50;49;00 - 00;51;18;16
Unknown
Yeah, yeah. So, okay, going back to a point you made earlier on the hiring side, right. I talk to one another CTO, Kyle, building out like an AI tech stack for brick and mortar SMB, but but less about who he's building for and more about how he builds and realizing one that because he's been able to build modular lead because of the tooling today, he's he's actually looking for more designers to try to get more of that customer experience side of things.

00;51;18;16 - 00;51;49;01
Unknown
Right. And he's totally changed his perception on how to build an organization in an AI native way. Like what? That means you're you're also looking at changing how you hire based on the current state, but it's less about being AI native and more just about like, how do we how do we approach maximizing and leveraging the tools to to solve the problems we we can uniquely solve, as opposed to just like being AI native for being eight AI natives?

00;51;49;01 - 00;52;15;12
Unknown
Worth forcing it. Yeah, yeah, yeah. So this is really interesting because, if I've thought a lot about this recently, the theory and I call it the theory because I haven't put any of this to practice yet. We don't hire people every single day, so don't get to try it that frequently. But the theory is, as these tools get better, the value of technical implementation decreases.

00;52;15;14 - 00;52;34;16
Unknown
Because I don't know if you've tried it, but I'm sure if you have written something in a programing language you didn't know previously, it's proof point that that used to take months of effort to learn the syntax, to learn the quirks, to learn how you know that language behaves at runtime and you don't need to do that anymore.

00;52;34;16 - 00;53;09;12
Unknown
I can take my entire code base and say, rewrite this in Python. And it would do it. Would it be perfect? Probably not, but it would do it. I don't need to know. Python to generate Python code anymore. And so the value of individual technical skills is decreasing over time. And therefore the value of problem solving business skills, customer empathy, product design, those things are increasing over time.

00;53;09;14 - 00;53;33;17
Unknown
Because if you think about the engineering life cycle, so much time and effort is spent on writing, testing and maintaining code, and you need specific people to do that. My program managers or product managers don't write code, and as a result, you need programmers to write the code. And I've always made this designation. There's a difference in my mind between a programmer and a software engineer.

00;53;33;20 - 00;54;00;23
Unknown
A programmer fingers on keyboard types and characters, but you have to tell them exactly what you want, and they implement to spec software developer, software engineer. I expect to put more effort into understanding the context of what they're doing and why. Be more active in the development of what they're building and the feedback of what they're building. And provide input there.

00;54;00;26 - 00;54;23;22
Unknown
And so over time, the need for a programmer is going to decrease. Your programing skills are going to matter less to me what languages you know, what technologies you use. I'm going to care less and less. I'm going to care more about architecture decisions, at least for now. In this window, the tools are pretty good at recommending architecture, but they still, in my opinion, make poor choices sometimes.

00;54;23;22 - 00;54;31;06
Unknown
So you need to babysit them when they're making big architecture choices.

00;54;31;08 - 00;55;02;24
Unknown
But like that layer and up is the skill set that's going to be valuable, and that's going to be less of people who are interested in the efficiency of the algorithm, tuning the database query, writing the new language, and more of the people who care about the end user, the business problem, the goals of the business, and how to achieve those, because those users will be able to accomplish programing tasks if they choose.

00;55;02;26 - 00;55;26;03
Unknown
Whereas the the hard thing to teach an agent or an LLM is the business reasoning and, customer centric side, that also might change at some point. You can imagine a world where every customer conversation you've ever had has been recorded and set in, and it can do a better job analyzing your customer's needs than you can.

00;55;26;06 - 00;55;53;24
Unknown
But at the moment, the most time consuming part is writing code. And that's the expensive part. From a company perspective, hiring engineers very expensive. And so the idea that when I'm hiring the team that will build our future product, rather than having the traditional 1 p.m., 5 to 10 engineers and that's your unit that goes and build stuff.

00;55;54;00 - 00;56;02;06
Unknown
I think that paradigm is going to shift to 5 or 6 PMS, 1 or 2 engineers.

00;56;02;08 - 00;56;23;14
Unknown
And and it probably can even move further up the chain. Because if you think about right now, PMS exist as an interface between developers and the rest of the business. At some point you'll have ops, people who don't need to tell a PM, this is what we need. They can just go self-service that themselves. Yeah, yeah. Instead of closing model.

00;56;23;16 - 00;56;50;17
Unknown
Is that kind of yeah. The closer you are to the business, the more valuable you are. And so that means understanding business fundamentals and what the business needs to succeed. And then also product things like what makes a product successful, customer empathy, all those kinds of things. What's, what's like your controversial position in this space that, that he'll, you'll die on?

00;56;50;19 - 00;57;12;15
Unknown
This is tough, because there's so many or, because you don't want to share. It's it's because it changes frequently. And I think, the the most controversial one is every time someone tells me that I need to drop everything I'm doing and go look at x, I product because it's magic and it's the future.

00;57;12;15 - 00;57;39;09
Unknown
And if you're not using it, you're falling behind. I get one of those calls probably every few weeks from someone being like, have you seen this thing? It's crazy. Is going to change the world. Have you fired your engineering team yet? And I say, I roll my eyes at most of those, and I think I think my hot take is that.

00;57;39;12 - 00;58;12;24
Unknown
I'm kind of scared even. Say this out loud. I think the a genetic AI wave we're in is overhyped. Not because it's not valuable, but just because people are making it something that it's not yet. I think it will get there. I don't know if it'll be in three months, six months, 12 months, five years, but it feels like people make it out to be something that it's not.

00;58;12;26 - 00;58;35;10
Unknown
And we've seen this so many times with technology where the hype train comes through, we bet the world on it. There's always a few who knock it out of the park, and so everyone elevates them and looks at those and says, well, like this company did it and they're crashing, so everyone should do it. And it's just not that simple.

00;58;35;10 - 00;58;57;12
Unknown
Like there's a reason that even though cloud native companies, that's something of like 15 years ago, at this point, everyone should be on the cloud. And yet you still have companies that are managing on prem resources and migrating from on prem to cloud. And if if it was true that when these paradigm shifts happen, if you don't get on board, you die.

00;58;57;15 - 00;59;17;19
Unknown
If that was true, it feels like a lot of these other companies should be dead, and yet they continue to exist. And I think the reason for that is their products are good and people still want to buy them. And as long as that's true, you can say these things like, if you're not replacing this with AI, you're falling behind.

00;59;17;19 - 00;59;40;08
Unknown
You'll be obsolete in a year. But if your product is good and people still want to buy it, it doesn't really matter how you get there. And you can choose if you want to not replace everyone with AI and say, you know, I was reading about, but Arizona tea, Arizona sweet tea, they, like, never, never raise their prices.

00;59;40;11 - 01;00;04;01
Unknown
And he's like, why would I do that? Like, we we make plenty of money, we're profitable. I pay my employees benefits. So like, why why do I need to make my team more expensive? Everyone would say, like, what are you thinking? Inflation bother them and care. And at the end of day it doesn't matter because people like his product and they still buy it.

01;00;04;04 - 01;00;21;17
Unknown
And so that's my I don't know, I love it. No, but I think I think in a way you are seeing a bit of the quiet part of the venture game out loud. It is. It does feel like in venture, at least for me, coming from oil and gas, it feels like an venture. It's a lot of talk, like all the time.

01;00;21;17 - 01;00;36;16
Unknown
I mean, marketing is like half of it, right? And it's saying. And it's not to me. At first it was like I was told, well, you got to kind of speak it into existence, right? Like you got to tell people because that's how you get excitement. That's how you get dollars in money flows to the story. And I believe all that.

01;00;36;18 - 01;00;56;29
Unknown
But it does feel like we're on another level. Like, like this is, it's constant and and maybe it's because just the nature of the society we live in today, particularly in the US, where everything has to be polarizing or it's not, you know, it's not worth the headline. And, and we're just so inundated with media sources and channels for us to consume.

01;00;56;29 - 01;01;18;18
Unknown
Yeah, that, that you do have to look for an extra edge, to that, but holy shit, do, do I feel like. Yeah, it's constant where I would say 90% of the things I hear, I you dig in a little bit. It's not like I have to uncover a ton of layers, but like, I lift one layer and I'm like, oh, no, this is Wizard of Oz.

01;01;18;20 - 01;01;35;21
Unknown
Like, this has nothing to do with what you're talking about. Which in a way, as a venture capitalist makes my job a little bit easier, but also a little bit worse, meaning like there's a lot more noise to dig through, but it's easier to find the signal when you do because you don't have to, like, do tons of diligence to realize, like, this is bullshit.

01;01;35;21 - 01;01;54;06
Unknown
And yeah, didn't take a lot for the House of cards to fall. Yeah, yeah, yeah. So I like, I, I, I agree that it can be a feel a somewhat controversial but I don't think I don't disagree, I guess is what I'm saying. I think this is something that we're going to struggle with a lot. And in an ever increasing amounts I just say an adventure as well.

01;01;54;06 - 01;02;17;17
Unknown
I think you probably get an even more magnified version of that. So we were fundraising and we got those questions constantly of like hat like, you know, when you replace your whole team with AI, what will your margin be like? It's like there's not even pricing on these services yet. Like literally all of the AI pricing has changed in the last year.

01;02;17;17 - 01;02;37;20
Unknown
If I had made a model of that, they would all be wrong. Yeah, yeah, they used to be unlimited plans doesn't even exist anymore. Right. And so and so like some of those things are definitely magnified. But even at its core, it's like you said, the Google article, you know, like you, you click into it and all of a sudden it's like, well, it didn't actually implement what a whole team of engineers did.

01;02;37;22 - 01;03;05;12
Unknown
It was close. And it's still really cool. And the fact that it got this far is mind blowing. But did it really replace a whole team of Google engineers for a year? No, no, they just did it. Yeah, yeah. And so will it someday for sure. That's just not today. Yeah, yeah. Man I, I, this is probably a whole nother conversation, but like the power law dynamic from the investor side.

01;03;05;17 - 01;03;33;24
Unknown
And as they seem like y'all just went through a fundraise and, and having a lot of these conversations with the VCs, but also high net worths and other types of investors. But like I, I see this constant tension between, hey, as a venture capitalist, like I do need those two to hit on those two that make it out of this wave for my economic structure to work, when the reality is that, like there are ten others that are going to be just phenomenally profitable businesses, but I don't care about it.

01;03;33;24 - 01;03;50;20
Unknown
Yeah, that are you, me, Arizona Tees and I don't care about because they don't grow as fast or they're not as big a market or they don't, you know, they don't hit on that magic source that caused them to be the winners of the the cloud, the mobile that I wave. And so that's my thing is you got to know your identity.

01;03;50;22 - 01;04;32;18
Unknown
Utility is not going to be a unicorn. And so we shouldn't be playing like we are playing that game. We don't need to make those kinds of extreme bets. It makes more sense for us to be conservative and try to build a company that is the most likely to succeed in some capacity. Yeah. If I was at a, you know, hyper funded AI product based startup, I would 100% be chasing the moon here because someone else is.

01;04;32;20 - 01;04;53;05
Unknown
And if you're not beating them, there's not a lot of room at the top. And so I think it kind of depends on what lane you're in, to kind of decide, you know, do you really need to lean into the AI first or the AI native side? So let me ask this stupid question. Then. Because I think I run into a lot of founders that I'm like, dude, don't take VC money.

01;04;53;05 - 01;05;10;25
Unknown
This isn't this isn't for you. Like, you are not the unicorn. And that's great. Like your life is going to be very different and much more relaxing. And for your, you know, in this in a lot of these cases for their personality type, it makes a lot of sense. But then the next question is like, okay, cool, where do I go for funding?

01;05;10;25 - 01;05;29;29
Unknown
You know, and and you guys have successfully run that process. Maybe not the tech speak specifically about utility, but like in general for the founders and companies out there that recognize, hey, I'm not chasing unicorn status, but I still need some funding to be able to get to the point where I need to get to. How do you approach that?

01;05;30;02 - 01;06;05;19
Unknown
Yeah, I think it's a good question because I think there's a couple considerations that go into it. One is how you build and grow your company. I think when you are chasing VC money, you're trying to expedite a timeline that otherwise probably isn't reasonable. You're trying to grow as fast as possible, even if it's pretty inefficient. And so I think one of the things is asking yourself, like, do you have the ability to slow down and build at an efficient and comfortable rate?

01;06;05;22 - 01;06;40;09
Unknown
Because if you don't, then you might not have a choice. It might be swing for the fence or don't swing at all. And then I think from there it's finding the people with either aligned interests by domain or, who are looking for types of investments to balance their overall portfolio. We've had luck with family offices, and we've also had luck with people who are particularly interested in antiquated spaces.

01;06;40;11 - 01;07;07;09
Unknown
Utilities are a great one. And so there are certain investors that aren't necessarily looking for exact business metrics and numbers. They're more looking for a investment in a space that they already care about for other reasons. Or they're trying to balance out risk they have elsewhere. But that also comes with you're not raising tons of money and you need to make it last a lot longer.

01;07;07;11 - 01;07;35;11
Unknown
And so the traditional, I think, raise tons of money, hire tons of people, grow revenue by explosive exponential numbers, or die trying has to turn into a more. Don't hire as many people do more with less burn, less cash and try to grow as you get customers rather than grow and then get customers. Got it. All right, man, I blew you up with questions.

01;07;35;11 - 01;08;05;27
Unknown
Let's wrap it up with. Do you have any questions for me? Yeah, I do, only because you went there first. The I have I have been curious from the VC side, it feels like when I am being pitched products from other startups, people are forcing AI into their product. And like I said, I'm a product person, so I always look at does it make sense for my user to put AI here?

01;08;05;27 - 01;08;38;17
Unknown
Yes or no? And there's some places where it's been delightful, using stripe for transactions. They have a great interface where you can tell it some data you want, and it will go write you a SQL query and pull that for you. And it's awesome. But there are also places where they don't use it and they could have ham fisted, you know, I everywhere, Microsoft, my old employer adding the actual AI physical keyboard button is like one of the most crazy things I've ever heard.

01;08;38;19 - 01;09;04;01
Unknown
And so I'm curious to hear your perspective when you're kind of seeing all these different startups. Does it feel like companies are forcing AI into their products because they think they have to have it? Or do you feel like there is a reasonable balance of, you know, consumer interests, customer interest and demand from the companies building?

01;09;04;07 - 01;09;24;19
Unknown
Yeah, I, I it's a great question because I don't think you're wrong in your observations there. Everybody is trying to find ways to to implement AI for different reasons. Right. And so go back to the, the discussion around would a venture venture backed company should look like. Right. And it's yeah, I'm trying to expedite a timeline.

01;09;24;19 - 01;09;46;15
Unknown
I'm trying to do X, Y or Z. Money flows into the story. Yeah, yeah. And so you have, you have a like a lot of well, I guess the way the way gets implemented is different depending on where you sit on that spectrum. So think of it, you know, more, more anecdotally, we have companies that are we're going through an exit process, right?

01;09;46;15 - 01;10;05;23
Unknown
They're like, hey, we've hired an investment banker. We want to go sell. Our investment maker comes back and says, you guys need to have AI in your story somewhere. Otherwise you will never be able to exit or you won't get the multiple or the valuation that you want. Right. And then let's just say like, Richard, we have like ten companies that have they're going through that process.

01;10;05;25 - 01;10;26;15
Unknown
I would argue half of them, all of them implemented some semblance of AI. But to your point, sometimes it works, sometimes it was like hamfisted it didn't really matter, right? The companies that were able to deploy it in a way that made a ton of sense for their industry, their customer base, their processes were like, why are we exiting now?

01;10;26;15 - 01;10;44;14
Unknown
This is working. Let's just keep growing. Right. And it's been interesting. Right? Because as a VC, you're like, well, shit, dude, you got to sell like we need liquidity, you know? Yeah, but but they're not wrong. You're right. Like it is helping. And then there's other companies where it's like, okay, a few press releases, whatever. Maybe you did some stuff, but it doesn't really matter, right?

01;10;44;17 - 01;11;08;16
Unknown
It doesn't accomplish anything beyond just saying I am leveraging AI to do X, Y, or Z for like the full company rebrands. I don't know if you've seen those, but we've had some products that we've been using where they rebrand the entire company. The one that comes to mind is, open phone. We use open phone. They rebranded to CLO just and now they're like trying to be an AI platform.

01;11;08;16 - 01;11;25;04
Unknown
And it's like, guys, you're a phone service. Yeah, a web based phone service. That's why we use you be a phone service. And if you add a bunch of other crap, I'm not going to use it. If you try to charge me for it, we're going to go use another phone like yeah, no, you're stay in your lane you know.

01;11;25;06 - 01;11;53;15
Unknown
Yeah. So the short answer is yes. Everybody's trying to implement AI and they struggle with understanding what that really means I think I think it's it's a useful exercise. At first you might say man that's like so dumb. Like this is so much noise. I, I and I felt that way for a long time. I guess the more glass half full perspective is like, you know, it's great because you start understanding where I should be used should be where, yeah, where it does work.

01;11;53;15 - 01;12;12;28
Unknown
And so it does provide some like testing ground for signal. Yeah. Like oh okay. Cool. It works there. I should be looking for something around there. Right. Now from a VC perspective in terms of sourcing and diligence and whatnot, like I think, I think the AI is less important, more. So it's like, hey, do do I have it?

01;12;12;29 - 01;12;41;09
Unknown
To your point, are you solving a real business problem? Do you have the subject matter expertise to be able to execute on this? Or do you have distribution, pre-built distribution that that allows for you to be able to scale faster once you do implement AI tools, or if you are building an AI native way. And so I, I, I think, yes, from like a company building standpoint from a VC side or sourcing and investing side.

01;12;41;12 - 01;13;00;28
Unknown
You know, I don't, I don't know if like, like the AI to me just becomes a bit of a tool. Yeah. Which maybe I shouldn't admit to because, like, on the fundraising side of VC, we're telling stories again. And so what's important to the story? Yeah, yeah, yeah. Hey, we're doing we're on the, you know, the bleeding edge of this wave.

01;13;00;28 - 01;13;26;21
Unknown
And here's all the things that we're doing. And this is why we're we're a better VC to invest in. Yeah. Then then the next one. Right. Like we know not only are we identifying and investing in AI native companies, but we're also deploying I started, you know, AI workflows into our own processes. Yeah. And so I, I, you know, I kind of feel like you're the, the best VCs probably are doing both.

01;13;26;23 - 01;13;50;07
Unknown
They are both aggrandizing how much AI is part of their processes and and part of what they're looking for. Yeah. But then they're only actually leveraging AI in the workflows where it makes sense to leverage AI. And at its core, when you evaluate a company you still care most about, do you have traction? Do you have distribution? Do people want to buy your product?

01;13;50;07 - 01;14;05;26
Unknown
Are you solving problems? I mean, for as much as like, what is it like it same. But it's like like it's for as much things change. They stay the same kind of thing. Like that's what it feels like. You still go back to the core, the basics. You just have to learn how to tell the better story around.

01;14;05;28 - 01;14;21;20
Unknown
Yeah. And so then that's when you get like, you get that resume effect where it's like, well, I know I have to embellish because everybody's embellished, right? Yeah. Like, yeah. But at the end of the day, you put me. You put me to work like you realize that I'm better than the next guy. Yeah, I do like that.

01;14;21;22 - 01;14;43;11
Unknown
Because I do agree it is important to participate in the excitement because it is exciting. The world is changing. Yeah, and in a cool way. While still maintaining that at the end of the day, you're trying to solve people's problems and and get paid doing it. Awesome. Dude. I love 11 on that. So Ben Kelly, CTO, Co-Founder of utility.

01;14;43;13 - 01;14;44;24
Unknown
Thanks, man. Thanks for having me.