417: The Best Tech Stack in the Age of AI

A couple of years ago, I tweeted that “the best tech stack is the one you already know.” To this day, this is one of my most resonating tweets. People keep bringing it back, and founders who've been around for a while seem to particularly agree with it. But AI changes things. Or does it?

This episode of The Bootstraped Founder is sponsored by Paddle.com

The blog post: https://thebootstrappedfounder.com/the-best-tech-stack-in-the-age-of-ai/
The podcast episode: https://tbf.fm/episodes/417-the-best-tech-stack-in-the-age-of-ai

Check out Podscan, the Podcast database that transcribes every podcast episode out there minutes after it gets released: https://podscan.fm
Send me a voicemail on Podline: https://podline.fm/arvid

You'll find my weekly article on my blog: https://thebootstrappedfounder.com

Podcast: https://thebootstrappedfounder.com/podcast

Newsletter: https://thebootstrappedfounder.com/newsletter

My book Zero to Sold: https://zerotosold.com/

My book The Embedded Entrepreneur: https://embeddedentrepreneur.com/

My course Find Your Following: https://findyourfollowing.com

Here are a few tools I use. Using my affiliate links will support my work at no additional cost to you.
- Notion (which I use to organize, write, coordinate, and archive my podcast + newsletter): https://affiliate.notion.so/465mv1536drx
- Riverside.fm (that's what I recorded this episode with): https://riverside.fm/?via=arvid
- TweetHunter (for speedy scheduling and writing Tweets): http://tweethunter.io/?via=arvid
- HypeFury (for massive Twitter analytics and scheduling): https://hypefury.com/?via=arvid60
- AudioPen (for taking voice notes and getting amazing summaries): https://audiopen.ai/?aff=PXErZ
- Descript (for word-based video editing, subtitles, and clips): https://www.descript.com/?lmref=3cf39Q
- ConvertKit (for email lists, newsletters, even finding sponsors): https://convertkit.com?lmref=bN9CZw

Creators and Guests

Host

Arvid Kahl

Empowering founders with kindness. Building in Public. Sold my SaaS FeedbackPanda for life-changing $ in 2019, now sharing my journey & what I learned.

What is The Bootstrapped Founder?

Arvid Kahl talks about starting and bootstrapping businesses, how to build an audience, and how to build in public.

Aaron: 00:00

Hey. It's Arvid, and this is the Bootstrap founder. A couple of years ago, I tweeted that the best tech stack is the one you already know. And to this day, this is one of my most resonating tweets. People keep bringing it back and founders who've been around for a while seem to particularly agree with that.

Aaron: 00:22

They've gone through the whole learning experience of trying new tech stacks only to find that investing a lot of time in a new technology often is not worth doing that effort. Like, what they wanted to do was actually trying to build a business quickly. So a quick word from our sponsor here, paddle.com, because they're a great example of a smart tech choice that massively speeds up and stabilizes the early days of building a business. I've been using paddle as my merchant of record, my payment provider for all my software businesses because they take care of taxes, currencies, track client transactions, update credit cards in the background so I can focus on dealing with my competitors and actually building the business and not having to deal with banks and financial regulators. So if you would rather just build your product, check out paddle.com as your merchant of record and payment provider.

Aaron: 01:12

Now recently a peer of mine asked a very interesting question that adds a layer to this idea of the best tech stack being the one you already know. They suggested that at this point the best tech stack is probably the one that the AI tool that you use for coding knows the most about. And I think that's a fascinating thought and I think it's kind of partially true or true ish because for many people who don't do their own coding anymore the best tech stack might not just be the one that they know best or know best from the past but the one that the coding AI handles pretty well. It does make sense, right? You use a tool might as well use what the tool thinks it's best but here's what I believe: while the original statement still holds true, the best tech stack is the one you already know, we have to think about it in its inverted form.

Aaron: 02:02

It doesn't matter if the AI model knows a coding language or framework the best. The quality of code the AI produces and the quality of code that ultimately ends up in your software product is still directly connected to our capacity of understanding and judging, reviewing and debugging that particular code. We still need to choose what we know best. Let me explain why. To make sense of this it's really helpful to understand just how these AI models are being built.

Aaron: 02:33

An AI model is effectively a token guesser that builds upon existing examples of the written word. Human written word or computer written word doesn't really matter. It has ingested hundreds of millions of documents, code bases, Stack Overflow conversations, blog posts, maybe even transcripts from IRC chats and forums. Like from there, it's built this internal model that makes it very capable of guessing what the next token should be in this series of tokens, which ends up being the piece of text that we get. Right?

Aaron: 03:05

The code it writes or the text that it writes to explain the code that is writes the comments, all that kind of stuff. It's just token after token guessed from this baseline of data that it knew from the past. And the data that flows into these models, the training data, for at least the first couple of generations was what people had publicly provided. Open code bases on GitHub, GitLab, SourceForge, and all these other places where people deploy code in blogs, in conversations, right? That's where it came from.

Aaron: 03:35

And the AI was trained on this data, which makes it obvious, I guess, that the AI has had the most exposure to the programming languages and frameworks that were the most popular at the point of its training. That's kind of the qualitative view, right? If it feels that this is the most popular, obviously there is the most stringent and coherent data on it. And from a quantitative view, the AI is also trained best on code that in aggregate has been most often used and most often written up until the point that the AI was trained on that data. So it's only logical that the AI has had the most exposure to things like JavaScript which over the last decade at least has seen this massive renaissance.

Aaron: 04:19

It's always been popular, but it's been really, really big because it works pretty well both in the front end and on the server. Thanks to things like Next. Js, hosted solutions like Vercel and React and React Server and React Native. And, you know, there's a lot of JS. And JavaScript became one of the most popular languages and always has been since its inception.

Aaron: 04:40

And the fact that it was pushed into the browser is the only language to work there. And a lot of people wrote code and published it and that made it available for AI to ingest. So languages like PHP, Ruby, maybe C and Java, equally popular frameworks with a lot of history in deployed code for decades at this point, they're pretty well represented internally in these AI systems. You can look at the statistics that GitHub sometimes releases about the kinds of coding that people do and have done in the past. They are kind of on a year by year basis.

Aaron: 05:13

But I would like to see actually what the historical distribution is. We might see a lot of JS, obviously, because it's very popular now. But I guess fifteen years ago, JavaScript was not as big. Like, back in the day, we just had gotten CoffeeScript, which was the first kind of transpiles into JavaScript thing or one of the first that then later led to things like TypeScript and so on, which are now very popular. But back in the day, JavaScript on the server was a rather new thing.

Aaron: 05:41

I'm just thinking like fifteen years ago. So in those days, people were writing a lot of Python and even a lot of Ruby on Rails was happening. And so those code bases may have been the most represented ones. And they're still on GitHub. I guess the older version of them definitely must be there, like, for each of these frameworks and for each of the applications written on it.

Aaron: 06:00

So, you know, there's a lot of data. That's what I'm trying to say. If you look for more esoteric code in a sense, like the more rare kind of languages, niche programming languages, things like Nim or Erlang or Smalltalk, well there might not be as much of it now because it's never been hyper popular, but there's still ample history for those languages. A lot of code was written, maybe not a lot, a lot like JavaScript, but still a lot. Might just not be as prevalent as other languages.

Aaron: 06:30

And that's assuming that all of this data actually made it into the training corpus. Maybe people were more selective and only trained AI models on, I don't know, just the popular things. The model can only know what it was fed, can only assume things then from the data it's been trained on. So when it comes to making a technological choice based on what the AI understands, well the answer is actually quite surprising here. You could probably use everything that has existed in the past, where there's a written record for code to be training material for an AI.

Aaron: 07:02

If we can be sure that the current AI systems have been trained on pretty much everything on GitHub, then you could use any technology that ever had a trace on GitHub and the AI will probably be fine with it right out of the box. Now if you go with JavaScript, PHP, Ruby, Python, those kind of things, you'll get very good results by default because the training data has been so plentiful already. But even if you go with Elixir, Scala, Bash, those kind of things, you still have a pretty massive data set for the AI to draw on. The problem might come with languages or frameworks that have been created since the AI was baseline trained because the AI only remembers what it was trained on in the past. We have AI systems that constantly create the new training data and pull it in without human intervention, you kind of still have to wait for a new version of an AI to be released that may or may not incorporate new features by default.

Aaron: 07:58

Now that obviously is just the AI system. If you start thinking about the extensions of these things, of a sudden the problems that I'm talking about are really not problems anymore because the MCP, the model context protocol, and the integration of tool calling in AI systems makes all of this somewhat moot. These tools are able to ingest the full documentation, any version of any framework and allow an AI system to update its knowledge on the fly as it is working on a coding problem. There are MCP solutions that pull in docs for every single library that you use. It's really cool not just library even the frameworks that you have that have going on.

Aaron: 08:37

So this means that if you have a proper MCP system that pulls documentation for any programming language that you use, that any sufficiently capable AI coding system, and I'm talking about things like ClotCode or Codex by OpenAI or Cursor or something like that will be able to code in the language you choose. So ultimately it does not really matter which framework you choose for your coding projects. The AI system with proper default choices or with proper instrumentation around these non default choices will be capable of building this tool for you or building anything you wanted. So for some people this might mean they think okay cool if that's the case well then I'm just going to build this project in whatever cool language I think is perfect for this. Maybe you're a JavaScript dev and you've heard that you really should write everything in Rust on the back end because memory safety, performance, that kind of stuff.

Aaron: 09:31

You could probably instruct your coding AI system to build a back end server in Rust and it very likely would be able to build code that actually works. But this is going to be code that you don't innately understand. You're going to be presented with code that your AI system will build for you and you'll have to say okay cool sure let's run this or wait this doesn't look right if you're not able to tell one from the other if you're not able to say well this looks good and this looks not good then you're effectively building a black box something so opaque that you just don't understand how it works. You only see that it does or it doesn't. And if it doesn't work you can only tell the AI to fix it.

Aaron: 10:18

You won't be able to fix it at least not without massive effort. But you can't tell it why it should fix it other than it not working or what it should fix or what it should fix it with and to what end that fix should be better or different from the current implementation. You won't be able to do this because you just don't know. And obviously, all these activities like debugging, expanding, manipulating code will likely be done by automation systems for many people in the future. So for most developers in the future, for many already at this point, these systems will do all their work.

Aaron: 10:51

But the capacity to see errors and to understand the infrastructure of a coding project is still the responsibility of the actual person instructing. Right? You are responsible for maintainability and the functional capabilities of your coding projects today and in the future. So I strongly, strongly recommend that even though we have this magical tool that can build anything in any language, you should still learn and deeply understand one particular programming language, one particular thing. Get really good at that.

Aaron: 11:25

For that choose one that has this vivid ecosystem that an AI can pull a lot of documentation from if needed but that still allows you to understand and choose a language that has a vivid community that you and an AI can pull conversations from and participate in conversations in and learn what the best practices are why they are like this right you want something with a strong historical backlog of existing projects That's kind of the Linde effect. Things that have stuck around for a long time are likely to stick around for a longer time. And AI systems are going to amplify this. The AI systems themselves right now are gonna create artifacts in particular frameworks, like things like Laravel or whatever that I'm using and then languages like PHP by creating new projects that people then upload to GitHub where then this goes into the next generation of AI models as training data. It's going to self amplify.

Aaron: 12:19

And if you want to choose a technology that you are versed in, then you are in the right path. Because if you understand any particular code that is written either by you or by a machine, then every potential commit that the AI suggests to you, you will know exactly what happens and you will maintain the quality and maintainability of the system because if you don't the AI will build what people who wrote code in that language thought in the past was going to be best. It's not what you think is best right now, it's not what you want to build and would build in another language, but what the collective implementation strategy of people in the past who used this technology back then and today looks like. That's what an AI will try to do. The AI is effectively going to build something kind of somebody else's code with your instruction and you don't want that.

Aaron: 13:11

You wanted to build your code with your instruction so for that you still need to know and understand the language. If you're on the fence right now whether you should use another framework or another tech stack that you know an AI will understand but you don't I think go back to what you understand and slowly learn this new technology. Do not throw completely new tech on new projects and hope that the AI will save you. I don't think that's gonna be the case. And even if it works, you still lack this understanding of it.

Aaron: 13:41

You need to be able to understand every single piece of code. You need to develop experience in potential scaling challenges, in esoteric edge cases, in handling small and frequent or large and infrequent data. You need experience integrating this tech with others and getting other people onboarded into this technology because your business will grow and you're going to be hiring and all of this requires that you have a solid foundational understanding of the technology that you're using so do not outsource this to AI the AI is only going to be as good as the other people who used that in the past and it does not necessarily map onto what you actually need. If you can't reliably instruct your AI with these things it will build something that somebody else would build not something that you would build and that doesn't work if you want to build your own business. AI's impact on tech stack choices is pretty massive because you can do everything, but you should be highly restrictive with these new technology choices that you outsource to AI because that is outsourcing control.

Aaron: 14:38

And that for founders like us ultimately is outsourcing ownership. Because if you can't fix it you have to hire somebody who can pay money for this if you can't fix it you have to rewrite it in the future and that's going to cause financial stress to your business and if you're going to have to rebuild it in a way that you understand then in the end, when you sell it, it's gonna be complicated because now you have this kind of Frankenstein code base or you have to stop using it altogether because the project doesn't work and you have to just turn it off because it spirals out of control. And you don't want that. So do not make AI centric tech choices. Make you centric tech choices.

Aaron: 15:14

The best tech stack is still the one you already know, even in the age of AI. And that's it for today. Thank you so much for listening to the Butcher Founder. You can find me on Twitter at avid khl, a I v I d k h l. If you want to know what everybody's saying about your brand, your name, your products, and over 4,000,000 podcasts, podscan.fmtracks mentions of anything you want to track in near real time with a very powerful API that turns podcast conversations into actionable data.

Aaron: 15:40

And if you're hunting for your next business idea, you can get that delivered fresh from the podcast world at ideas.podscanfm, where we extract the best startup opportunities from hundreds of hours of expert conversations every day. Thank you so much for listening. Have a wonderful