The Bootstrapped Founder

Happy New Year. Hey. It's Arvid, and this is the Bootstrap Founder. This episode is sponsored by pedal.com, my payment provider of choice. If you are in a software business and you don't want to deal with the intricacies of sales tax or having to integrate every single money related thing yourself, I can really recommend it.

Paddle made this all super easy for me, and I use this in lots of my properties. So go check it out. It'll make operating your own software business much better. Last week, I did a recap of my 2024. And now that we're starting a new year, let's look ahead at the opportunities and the challenges facing the software business world of 2025.

I want to share my insights into the SaaS and software engineering landscape. Just from what I learned by being a SaaS founder. Look at all the emerging trends and patterns that I see and explore in my own work. And most importantly, how to leverage them. Whether you're just looking for business opportunities for yourself or just trying to stay ahead of the curve.

There's something to learn from all of these things. The biggest impact on how we will build and run businesses in 2025 will definitely come from agentic AI and I know this is a buzzword that everybody is talking about AI agents and I know it's annoying and this is probably one of the 20 times you've heard this already, but it doesn't feel like a revolutionary shift from what is already out there. Especially if you're a software developer and you're used to AI coding assistance and that kind of stuff. But I believe we're about to see a major leap here in just the capabilities of the systems and what comes out of this. When people discuss Adjentic AI, they often use like these trivial old school examples like ordering flights or pizzas.

And these examples remind me of the 1st Bitcoin pizza purchase, which is almost a meme at this point and Bitcoin itself recently crossed the $100,000 mark and had made the whole thing, the purchase of that pizza a half a $1,000,000 pizza in retrospect. Which is fun. And it kind of gets you views if you talk about it. But this pizza example undersold Bitcoin's potential clearly. And reducing AI agents to simple task automations also misses their transformative power.

So let me share maybe a more compelling example that I recently encountered on the Ted Talks daily podcast that I occasionally listen to. Maybe not every day, but occasionally. And that day I listened to an episode about AI's application in biotech. Specifically AlphaFold protein folding simulations for drug discovery. It's very specific but that made it so interesting because currently pharmaceutical companies employ teams of researchers who might spend years studying a single proteins properties just to get it right, just to understand it.

Like there are whole PhDs written about one individual kind of thing, one protein and its properties. The podcast revealed that AI can accomplish what takes a human years, the same task, in minutes now. And this isn't just about speed either, it's about scale. Because when you can replace years of human research with minutes of computation, and then parallel process thousands of these tasks simultaneously across these GPU clusters that everybody is building, well, the implications all of a sudden become staggering. Those 23 drug researchers in a company, they aren't just being assisted by AI.

Their entire workflow is being transformed. And the AI systems never sleep. They work thousands of times faster and they continuously self optimize through incredibly tight feedback loops. Imagine this with a person, right? Takes you a year to figure out the protein folding and everything you learn in between you can apply to the next year.

Well, an AI can do this within minutes and learn and learn and learn much faster. This computational power enables a fundamental shift in approach instead of developing generic drugs for broad populations like we have right now. We can soon imagine and maybe expect hyper individualized medicine. Imagine AI agents designing treatment specifically tailored to you, to your individual biochemistry and the illness you may be having running thousands of simulations in parallel for millions of people like you at the same time. The pharmacy of the future might not even stock generic medications anymore, but rather synthesize personalized treatments on demand and on the fly.

That is crazy to think about just coming from a world where this does not exist, but it could well be a reality and agents can do things like this. The transformation here is not limited to biotech. As a developer, I've already seen how AI has changed coding my own coding workflow. Tools like Cursor, GitHub Copilot. They have eliminated the need to manually translate my mental models into code line by line.

That era I feel is gone and I don't think it's ever going to come back unless you count artisanal coding or stuff like that. And there will be room for that. There will be a place for this. But for a capitalist society that is driven on efficient process and creating things cheaply quickly that era is over. We'll never again write code without AI assistance unless we have an outstanding reason to do so.

And what's particularly exciting is the decreasing size of capable language models. When we started out with this whole large language model thing like a year and a half ago, which is also crazy that this is just like a year and a half old really as technology hitting the scene, Everything was massive, right? The Chechi PTE was sitting on a massive closed non public model and there weren't any models to play with. And then all of a sudden Meta made the choice to release the llama model and Mistra released Air model. And there are so many smaller models that just started proliferating and people started training them and learning from them and building new ones.

And all of a sudden they're becoming smaller, but they retain the power of the big ones. It's really cool. We're moving from this cloud dependent solution that we used to have and still do in terms of scale for many processes to local processing, which becomes easier and easier because GPU is now become a thing that many hosting providers think they might need to attach to their servers. All of a sudden you can run a GPU based model on a web server because they offer these now and the prices are coming down as well because scale hits, right? Economies of scale.

So when AI first hit the scene we all relied on these big APIs OpenAI and JGPT and even Anthropics Claude still big APIs. Now with tools like llama. Cpp and olama tools that I use in my product myself you can run these sophisticated models locally on your own hardware and under your own control. This shift towards edge computing brings greater privacy, like nobody gets to see your data. You can do tenant isolation, like your model is not using your other customers data when you want to do something for one particular customer.

The data security improves with that and independence from cloud platforms and how quickly they can pull the plug also increases. That's a net win for founders, for people who want to have control over that technology stack. And for SaaS providers, this means a fundamental shift in customer expectations too, because if everybody can do it, then people will think you might want to do it anywhere and they will expect it in any product. You're no longer just providing tools that facilitate human work that make it easier. You're expected to provide intelligent agents that actually complete work and entire workflows, entire processes.

We've been talking so much about jobs to be done in the founder world and how you want to find a job to be done. And then hopefully give the person doing the job all the tools they need to get from A to B, right, to do the job. But now we're actually looking at jobs to be done and think, well, can we do the full job and can we have an AI do it? That is novel. Like that didn't used to be a way of building a business, but now it actually might be a bug tracking system that just won't just identify issues anymore.

It will have to actually resolve them, fix them. If you look at Sentry, that's exactly what they're building, right. They've been doing this very interesting thing. First, obviously they collect error messages, right. They are a bug tracker, an error tracking tool And the initial step was to kind of combine all this external information, like what deployment was it?

And when did the bug start? When was it fixed? Like what is responsible? And they've been doing this for decades. Like they have code and deployment states and code changes and bugs and bugs that were fixed and data about this in probably the billions of examples.

And now they started building a system that can actually start fixing these things for you that create a pull request into your repository with what the system thinks is the best fix. Well, guess what that is being trained on, right? Your existing code base plus all the changes, all the fixes of everybody in the past. This system is now built to eventually probably automatically resolve them. So you don't even know anymore.

So the pressure to expand from point solutions to comprehensive workflow automation will be intense for founders this year and for many years to come. This is not gonna go away. This is gonna get even harder and the shift is happening already. Recently my friend Tyler Tringus shared how he built an entire SaaS product in 4 hours using AI And he's a capable founder, he's a developer, he can build these things. So what a capable founder, developer like this can do today, anyone will likely be able to do next year.

Even less capable founders and less capable developers, the barrier to entry for creating software products is disappearing. Even the job of the software engineer is threatened by this. Not the people's skill set and the capacity for them to find a job. I think that will still be there. It's just gonna be something different.

You're gonna be an orchestrator more than an implementationalist, right? That's the difference now. The AI agents will do the work for you and you just check if what they're doing is good, not right. Correctness is not the problem, but good. Like something that is qualitative, something that you cannot infer that judgment will still be the task of a human being at least for the time being.

I don't know what I'm gonna say next year maybe there will be AI smart enough to actually do this already, but let's just say the execution will be AI ified in almost anything that has to do with computing computation. So you better keep your eyes on that and look for automation opportunities. Because people will expect them and they will pay money for it. And counterintuitively here, as automation increases, the human element becomes more crucial. Right.

Our capacity to judge. In a world where AI can spin up a competing product and ours. Well, trust and human connection actually become key differentiators for you and your brand. I've seen founders succeed by building in public and sharing their journey and maintaining genuine relationships with their customers against an onslaught of competitors and copycats and clones. People still seek the original because that's a person they can actually trust, not a nameless something that is just churning out these copies.

Your about page, your public presence, your authentic voice, these become competitive advantages now. When customers can choose between dozens of functionally identical products, they will choose the one they have chosen in the past, they will keep choosing the one with a real relatable human behind it. As the software becomes increasingly automated, and that is what we will see, being visibly human might actually be our strongest defense against commoditization through AI. And of course, this too might change if AI agents start handling software acquisition. If actually a company has an automated system negotiating prices and features while customizing solutions in the background, I can see this.

I can see big companies actually trying to find an AI agent to buy their tech for them to negotiate you into oblivion. So we will probably, as software founders, have sales AI agents that have to then fight with these purchasing agents and have, like, millions of messages and, negotiation tricks exchanged per second or something. It's gonna be pretty weird, but stuff like this will exist. But for now, at least for this year, human relationships remain crucial to business success. And whether we embrace it or not, AI will permeate every aspect of software business in 2025.

It already is. It already is so present right in our community clearly. But if even our boomer parents talk about chat gpt and use chatbots, like how could it not be in the whole world of software businesses and everything they facilitate? So our customers' expectations for all generations will be shaped by the best AI powered experiences out there. And the question isn't whether to adapt, but how to do so while maintaining the human touch that makes our businesses unique.

So, yeah, that's 2025 for you. At least my interpretation of what is to come. Thank you so much for listening to the Bootstrap founder. You can find me on Twitter at Avid Kal and you find my books on my Twitter cluster too. If you want to support me in the show, please tell everybody you know about podscan.fm and leave a rating and the review by going to ratethispodcast.com/founder.

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 will help the show. Thank you so much for listening. Have a wonderful day and bye bye.