Does Building SaaS Still Make Sense in the Age of AI?

The speed at which AI is taking over parts of human work is growing exponentially. What only two years ago I treated as support for writing and refactoring code, or for generating inspiration for visual projects, has now become an autonomous tool that does those jobs surprisingly well. With AI tools this advanced, does building SaaS still make any sense at all?

Today, creating an application requires almost no programming knowledge. Any logically thinking person who spends a few days studying the technology and the process of building apps, and makes an effort to dig a little deeper, can use AI tools to create a fully functional product. And sure, a layperson may not be able to build a large commercial product right away, but a good share of the people interested in the subject will certainly be able to solve their everyday problems with software, making their daily work easier in the process.

Let's also note that right now we have the worst version of LLMs that will ever be available. And even at this stage they make it possible to copy almost any digital product, depending on how deep the pockets are of the person who wants to build their own version of some app.

I'll try to keep this article as concise as possible and cover the basic issues in this area, and to wrap up I'll present my own products as examples. We'll start with the pessimistic scenario. Or maybe the optimistic one, depending on where you're sitting. After all, the possibilities AI gives non-programmers in building applications are, hypothetically, a utopian vision that just a few years ago was something abstract.

The death of subscription-based apps

In my opinion there are several main reasons behind this thesis. The first and most obvious is the development of LLMs. And in this case it's not just about advanced use in a terminal, on Linux, by some nerd in a basement who wants to build a new superintelligent model by iterating in an infinite loop. Sure, the very fact that ever better models are being built and used by specialists matters, but what I mean is more about what the whole environment around this endeavour looks like. The desktop, and even mobile, apps of the industry's biggest players (Claude, ChatGPT, Gemini) are child's play to use. They require absolutely no IT knowledge. Even if we're building a more complex product, such as a game, an application or a system for analysing specific data, we'll be led by the hand. Giving models access to operate our computer, or handing them API keys and thereby letting them edit and manage the specific tools we subscribe to, gives them complete autonomy in carrying out our instructions and implementing specific design decisions. That fact alone allows LLMs an unprecedented degree of self-sufficiency.

On top of that, the ecosystem around specific models is developing at an alarming pace. The apps that give access to the models keep getting new dedicated features: design, multitasking on projects, coding, generating videos or graphics, successfully replacing tools such as the Adobe suite, Figma, Notion, VS Code — or rather, in that last case, the general ability to use programming languages.

What the sceptics say

Now, there will surely be plenty of seniors with 20 years of experience in each of these fields who will challenge the paragraph above by saying that things generated with natural language will never be as good as what an experienced software architect creates. Or that even if they are, sooner or later they'll break down in a way nobody will be able to fix, or that they simply have no soul and something is missing. Sure, I'll agree with many of these arguments, but the egos of people who have spent the lion's share of their lives learning a specific skill often obscure the full picture for them. From a pragmatic point of view, not everyone needs a perfect tool that is polished in every detail and works like a Swiss army knife. Often we simply need a tool that solves a specific small problem in the shortest possible time. We put the hell of perfectionism aside, assuming we'll gradually refine our tool, but what we care about most is that it quite simply works and helps us in our daily work.

From search engine to AI

I think this is the crux of the matter. Just a few years ago we had a problem to solve. We typed it into Google and often got several cloud applications (SaaS) that could solve it. We worked out whether it paid off and whether the cost of the subscription would really be recouped by speeding up our work, and that was that. Today we type the same thing into AI. Often, right inside the app that runs the model, we get a ready-made Excel file with a specific formula or a simple HTML file with JS that easily produces the effect we're after. If the app window can't generate a given result or file, or it falls outside the scope of the dedicated tools in that app, we get specific instructions, move to the terminal and within a few hours can build our own simple CRM or another data analysis application.

We're dealing with two fundamental factors working against SaaS here. The first is how advanced the apps running LLMs have become: they can show us ready-made solutions to our problems directly in the browser window, as micro-apps, visualisations or other ready tools. The second is generating a specific tool for us when the scope of a given model app can't solve our problem directly. Exactly what I predicted a few years ago is happening. In the near future anyone will be able to give a model a few sentences about exactly what they need, and the output of that input prompt will be a finished product that only we, or a small group of people, e.g. our small team, will use.

Tailor-made software

There is also the aspect of laser-precise fitting of a solution to our specific needs. Most web applications solve a problem in a general way, at a macro scale. More than once I've had to rebuild some all-in-one CRM monster specifically for a given client, because the system's assumptions met the client's initial expectations, but at a more precise level we had to, for example, switch from pieces to cartons, rework the way costs were calculated in the system, or change some specific logic. Today, LLMs with access to our specific data create tailor-made applications that solve the problems of very narrow niches — niches that a programmer building a product for a particular industry may have no idea about.

So in this case, not only can the client solve their own problem with software, but on top of that the product generated by the model will fit their needs more precisely. And it will be created for a small fraction of the cost of software built by a team of designers and programmers. Sad, but true. If someone denies this picture of reality, they are probably not so much a sceptic as an opponent of using models, or they work in such a narrow and technologically advanced niche that a lot of water will indeed have to flow under the bridge before those problems are effectively solved by LLMs too — but more on that a bit later.

To sum up the part about the imminent decline of SaaS building by large IT companies: the situation looks rather grim, but every coin has two sides, and in the next part I'll try to show the bright side of the current state of affairs.

A renaissance of software building?

In the first part I presented a rather utopian vision, assuming that we live in a society of intellectuals, where everyone can skilfully use pragmatic logic and writing pseudocode for a new digital product is a morning warm-up before settling down to read Hegel. Unfortunately, or perhaps fortunately, the world doesn't look like that. Of course, the barrier to entry into software development has been drastically lowered. A person who has never had any contact with the IT world, and for whom understanding binary code in high school computer science classes bordered on a miracle, can still create something that solves a certain problem — but can they really?

I think there won't be as many such people as we imagine. Of course, the amount of software being built will grow. A good chunk of it will be total crap, AI slop, call it what you will — we could pile on plenty of labels here, dressing up apps generated by trial and error in ever new epithets. But what still remains is the will, the determination to reach a goal and the ability to think analytically.

Who knows "what to type in"?

I gained some perspective on this from a remark by one of my clients — a fairly elderly but still very inventive and, in my view, exceptionally creative person, who can see things through to the end like no one else. The remark went roughly like this: "Michał, we've got this AI, it already does almost everything, but sometimes I don't even know what to type in". And that's exactly the point. The edge in the technology race, or the ability to put certain ideas into practice, will belong to the people who know "what to type in". People who take even a little interest in a subject and devote themselves with focus to the process of creating, planning and analysing, so that within a few iterations they get exactly what they asked AI for.

An opportunity for the ambitious

This is where a big opportunity opens up for a certain ambitious group of people. For the inventive and intelligent part of society that has always had some great idea for an app, a game, an animation, anything. But they always lacked the time because of, say, a full-time job, and breaking into certain industries was a Sisyphean task: having rolled the boulder to the top, they would suddenly see a new, even bigger rock formation emerge from behind the summit, which instantly cooled their enthusiasm and showed how far they still were from a finished product.

I can give my own example, staying within our main thread, i.e. SaaS. UX/UI, Figma, HTML/CSS/SCSS, then JS — I already think I've made it, because at university I can build a finished application, but it turns out I haven't, because the front end alone isn't enough. Then comes the tedious learning of more technologies: MySQL, PHP, WP — because maybe it'll speed up the work a bit. Phew, finally the front end works, connects to the back end, something gets saved in the database and we've made it. But no, because maybe it's better to use more modern technologies? TS, Next, Angular, React, Node, Supabase... and so on forever. And this is exactly where a certain window opens.

People who have a certain skill set and range of knowledge can, as never before, carry out their ideas very effectively and much faster without a large team of programmers. Exceptional individuals are now able to create unprecedented tools, applications and systems in far less time. Whether that's really the case, and whether it always looks so rosy, I'll write about in a separate article. But in an idealised world we can hypothetically assume that modern AI models really do give us that chance.

The idea matters more than the technology

We want to create something — we don't have to spend dozens, or rather hundreds, of hours learning C, Java, Unity or Blender to make a simple Flappy Bird-style game. AI will create it for us or lead us by the hand step by step. We can generate a finished product and then, through reverse engineering, learn its details and edit and adjust it step by step until it precisely meets all our expectations. That gives us an amazing opportunity to create things that just a few years ago were out of our reach in terms of both time and technology.

It's the same in the world of SaaS. All the technologies I wrote about a few paragraphs earlier may matter far less than an innovative idea and a very good understanding of a problem in a given industry. In my view, AI models have shifted the weight of responsibility from the technological side to the problem-and-logic side. This means that the idea itself and the app's assumptions are now far more important and crucial to delivering projects than the design or the technology itself, where an LLM can do the lion's share of the work.

Answering the question "Does building SaaS still make sense in the age of AI?"

Here, too, I think the answers may differ depending on the industry, the product, the problem being solved and its scale, but all in all I believe the answer is yes, and I'll try to show it with a few concrete examples:

1. Products that already exist, but can always be done better or cheaper. This creates a huge opportunity for people working in a given industry. Knowing it very well, and knowing which SaaS services are available for a particular market segment, they can create a better app that solves the problem. Something cheaper, better suited to the user's needs, or simply simpler and more intuitive to use. That makes it possible to take over a share of a specific market as far as software is concerned. Alternatively, you can become an expert in some field as a hobby, or create a dedicated product for the company you work for, which gives you a more responsible role in the organisation, increasing your company's profits or simplifying its business processes. For example, even though there are many great design systems, I've been building my own with the help of AI for a while — https://www.prfct.dev/. It may not be anything groundbreaking, but it's a fun hobby project for me that I'll be able to share with the community at some point.

2. Building complex tools that aren't just wrappers around AI models. I keep coming across apps, aggressively promoted through ads or through spam flooding forums and discussion groups, that are nothing more than LLM wrappers with a nice UX/UI. Typing the same thing into any "chat" would get you a similar result. The app maker buys tokens, plugs in the API, adds their own margin and has a simple money-printing machine. Sometimes they equip it with a SKILL or an extensive prompt that moderates the user's input, and voilà — we have a hook for unsuspecting users.

   Things change when the person building the product really knows what they're doing. Building the right prompts isn't easy. The app draws heavily on MCP servers, often combines many AI models dedicated to specific tasks, and properly configured agents work independently of each other, giving an incomparably better result than typing our need into ChatGPT or Claude. That's exactly the motivation behind my app brandmaker.studio, which aims to replace an advertising agency when it comes to building a brand and its complete branding. I'll describe the work on it in an upcoming dedicated article.

3. Building innovative tools in real time. The AI revolution will lead not only to new methods, technologies, or even whole industries and new fields, but it will also generate new problems to solve. To a large extent these problems will be solved by LLMs, but certainly not all of them. A creator will be able to spot a specific issue that needs simplifying, or a problem that calls for some solution, and within a short time build an app or tool that solves it. In such cases, a very precise and advanced knowledge of the industry will matter, and so will time — the competition never sleeps, and as history has shown us, being first at something is often more important than being better but second.

Summary

These are only my musings and some thoughts. In the world of creators, programmers, designers, inventors and other people for whom creating products matters far more than consuming them, there will probably be as many supporters as opponents of my modest, subjective theories. The truth is that it's the market that really verifies hypotheses, and speculation about the future has more than once been judged brutally by time.

That doesn't change the fact that we live in times of a technological breakthrough on an unprecedented scale, especially in terms of the pace of progress. The human brain can't keep up with it, and a kind of disorientation and chaos accompanies a growing group of creators.

The conclusion, however, is that we can't give up on constructive, innovative thinking. If not SaaS, then maybe something else. Maybe dedicated small products for individual companies. Maybe solving complex processes and problems in yet another way that is still to come. Time will tell. Still, continuous development and the pursuit of innovation may turn out to be the key to creating something that ultimately becomes a breakthrough solution, giving us the long-awaited satisfaction of having built something that has a positive impact on the world around us.