The Bootstrapped Founder

Hey. It's Arvid, and this is the Bootstrap Founder. This episode is sponsored by paddle.com, and today, we'll talk about habits. Well, habits, actually. The way people approach work and the expectations that they have for the tools that they get to use, but I did promise Hobbits.

So little story. When Peter Jackson was brainstorming the Lord of the Rings movie universe, like, long before the first cameras were rolling, 2 amazing Tolkien savvy illustrators set in on those meetings. That was

John Howe and Alan Lee. They had illustrated a lot of the books before, and Jackson was quite inspired by their work. So he got them to work on this initial brainstorming stage.

And Jackson would explain a scene, explain the world, what was where and what the buildings were gonna look like and all that. And the artist would sketch as he talked. So when he was done explaining, the 2 artists would hand over their quick sketches, and that allowed this director to see what he had just envisioned, but on paper. And that was great for him because he could immediately see the results of his brainstorming in a shape that he could share with others. And until now, you needed extraordinarily gifted experts with years of experience to do this kind of work effectively.

But, I guarantee you, someone out there is currently working on a generative AI tool that takes and visualizes ideas like this as they are spoken in real time at scale as an app. And on the off chance that no one is working on that just yet, maybe you should. Because in the months to come professionals in all kinds of industries will expect this kind of tool to exist for their own use cases. The expectations around AI powered software have shifted dramatically here. I've witnessed this transformation firsthand while building BotScan and it's fascinating to see how this relationship that we have with AI tools has evolved over just the last couple of years, couple of months even.

Remember when AI was this mysterious thing that only existed in these specialized expensive tools? But then it's just 2 or 3 years ago. And back then AI meant stuff like machine translation or predictive analytics or even neural networks churning through massive datasets. Stuff that barely anybody could afford and that nobody really understood. It was specific, it was limited and usually hidden behind complex interfaces or databases that nobody had access to.

Only the people with

a lot of money, a

lot of funding. The tools themselves were compartmentalized. You had one for translation, another one for data analysis and yet another for predictions. But, look at what's happening right now. I opened Ahrefs the other day, my go to SEO tool.

I noticed something very telling. Their AI content helper, the link to that part of the tool has claimed the prime spot and the navigation. And it pushed traditional features like the site explorer and the keyword explorer towards the back, which is I think the mainstay of this tool. Like the the things that people actually use most of the time. Or so I think.

Right? I don't think this is just a UI decision. I think it reflects this fundamental shift in how we expect software to work. It's not just a thing that we use and then get some results maybe. It is something that does work for us.

And that's why these AI parts of the tools that we use have become front and center because the companies behind these tools, they see the adoption, they see what people really want. And through my work on PodScan, I've discovered something particularly interesting about the capabilities of these systems. While we often worry about hallucination, which is often happening in creative tasks and generative AI. I have found AI to be remarkably reliable in analytical work. When I use AI for my own question answering and my data extraction, which I do a lot on podcasts, it is impressively accurate.

And the key insight here, well, AI excels at reducing complexity, but it struggles with expanding it. It has a hard time coming up with really cool interesting insights, but if the insights are already in the thing that you ask it to investigate, it is really good at packing them into something consumable and meaningful that may be even less verbose and more specific than the thing in the original data. It's like the difference between summarizing a book and writing one from scratch. AI is spectacular at summarizing and so so in writing one. And this observation leads me to envision, I guess, 3 distinct ways that AI is reshaping software tools.

And if you are a software founder or a founder in any kind of business, you might want to think about the utilization of AI for these 3 particular kinds of ways that people want to interact with it. The first one is AI as a guide. Imagine a sales pipeline tool that doesn't just show you how to use features, but it actually guides your strategy in reaching out to people. It might suggest, hey, you should probably get customer success to reach out to this other specific customer that we have for testimonial, because their use case would perfectly fit what your current sales conversation needs. The AI becomes less about interface navigation and like creating text for you.

It becomes more about achieving your actual business goals. Pulling data together and seeing where you can apply it more effectively than before. The other way I see AI reshaping tooling is as a validator. And that's kind of where this question answering data analysis angle comes in. Right?

That's where my experience with PodScan has been very enlightening. AI shines when it comes to verification and validation. While we often hear about, AI came up with random non factual stuff, That's just creative tasks for analytical work, fact checking, data verification, consistency checks. AI is remarkably reliable. Particularly, when it has function calling enabled.

I think most of the chat gpts and cloud services have something like this where you can actually ask it to call a function to execute some code, maybe to search the web. I think perplexity as a tool does precisely this. So having this meticulous assistant who never gets tired of double checking your work in their own work, Which the more recent reasoning models, like, o one and soon o three from OpenAI are particularly good at. Right. They look at what they created and they recreate it, they recreate it until they they find it very consistent.

And then they present it to you. They go through steps of reasoning. And finally, here's where things get really interesting in the third way that AI is gonna be interesting moving forward. AI can act as this constant background presence to any kind of work, be it coding work or writing work or operational work, because AI can constantly scan your business data for opportunities that you might miss. As long as you present that data and you gather that data and you capture that data.

Think of it as an AI agent that works while you sleep. For developers, this example is pretty clear. It could mean just an AI system that continuously analyzes your code base and not just finding bugs because that's what existing services do pretty well. You know, if a bug happens, you track it and it goes into Sentry or a system like this, but it can actually fix them through simulation and testing. I I think Sentry is a good example here because they have been pushing for something like this over the last couple months, I think.

Like, Sentry is a service where if your software runs into an error it sends the exception message and the context of it, like what code captured this, what the code was that didn't work, what the data was that flowed into the system, where the request came from, what the server was, all of this stuff. Right? So Sentry for years now has been collecting all of this data about all kinds of bugs in all kinds of software. They have thousands of customers with thousands of different code bases and many different programming languages, but they all run into errors and Sentry has been capturing this. So what Sentry has recently implemented is kind of a LLM based code fixing suggestion that would create a pull request with a potential fix for your code that you then have to look at and test and see if it works and then you can say, yeah, sure and it actually pulls it into your code base and deploys it.

Because one of the wonderful things about Sentry is that it also tracks your deployment. So it knows when a certain error started to happen it can track them back to the certain deployment and the code changes within. It's really cool. And they have all this data lying around. So obviously they can train models on this massive amount of data that they have.

And in building these models to then solve these problems, I think the next step here is that they will automatically simulate a system where your code is running, where the fix is running and they will check the outcomes from the system with the error compared to the system without the error and hope that the fix actually works. That's kind of how the system would constantly simulate new changes, new fixes, maybe new error States that you may not even have recognized before and create fixes while you sleep or while you work on something else. I think this is what AI agents are particularly interesting for. They are so useful to get work done that you wouldn't even think of. And I think the implications here go far beyond tech.

Imagine an AI that notices that you haven't spoken to a key customer in a couple weeks and remembers that they usually attend a conference every year because they've been tweeting about it every single year. And suggest that you should schedule a meeting because you're also going to this conference. Or one agent that recalls a customer mentioned launching an article in April back in October last year when you last talked to them and reminds you to follow-up in March about this article and how great it was and how you shared it with your own customer base. Right? These little things can be done by systems that act as kind of an opportunity scout for you.

And what fascinates me is how AI is evolving from being this mere text and image generator to becoming this ambient intelligence that enhances every aspect of our work that we let it be enhanced by. In my experience building and using AI tools with pod skin and all the other things that I'm doing, I've seen that the real power is not in flashy features, but in this quiet persistent assistance that makes everything that we do a little bit better, a little bit easier, and a little bit more impactful. So the future of AI in software isn't just about chatbots generating images. It's about having this intelligent presence that helps you source opportunities and it guides you through processes for a better outcome and it validates your work along the way. It's becoming this background rhythm of modern software and it's orchestrating improvements in ways that we're only beginning to understand.

But people are starting to expect it. People think that this should be part of whatever tools that they use. And this is why I talked about this today, because I think over the next couple of months, definitely over the next couple of years, we will see this become just an expectation that you will have to fulfill. So as somebody building in the space, I can tell you we're just scratching the surface here. What's possible when AI becomes less of a feature and more of a foundation, and it's on us to make that a reality.

And that's it for today. Thank you for listening to the Bootstrap Founder. You can find me on Twitter at avidkahl, a r v I d k a h l, and you find my books on my Twitter course there too. If you wanna support me in this show, please tell everybody you know about podscan.fm and leave a rating and review by going to rate this podcast.com/founder. It makes a massive difference if you show up there because then the podcast will show up in other people's feeds, and any of this truly helps the show.

Thanks so much for listening. Have a wonderful day, and bye bye.