Building custom software with AI
I think the good way to build any SaaS right now, or any software for that matter, is to just start from the exact things that you need. Much of software has become cluttered by trying to make it useful for a “generic” audience or “most” of the people. But people have different needs. While many users might need a core feature, some features are simply more important to certain users than others.
With AI, we can create truly custom software. This enables people to build tools that fulfill their specific needs. Sure, you will have to figure out a lot of things as you realize what exactly it is you need, and maybe a generic software can just give you that. Finding a tool where It Just Works is what will really help simplify things in the long term.
But in some cases, the right software is just not there. Or maybe it’s too pricey, or it just gives you too many unnecessary features. Even if you get a good piece of software, you will have to learn way more than you would want to—unless it just works perfectly out of the box. For any generic use case, a product has to support a bunch of things to fit a lot of people’s requirements. Achieving that “it just works” level of product quality is incredibly hard.
That is why having AI build the exact things you need is a great enabler. It will always just work for the cases you actually want it for. It will not be built for things that you don’t even need.
Like right now, I am writing this blog using just voice typing with an AI agent that pushes to a commit in a GitHub repo, which is then displayed on this website. It’s very easy and it just works for me. I can write as many articles as I want in a day, and it doesn’t take away from my schedule. I can talk for a few minutes, and an article is ready in a state that is acceptable.
Perfectionism is the enemy of the great. I could make this article 100% perfect, but it would take a lot of time. The marginal rate of return on the invested time would severely diminish if I tried to make it perfect. But if I just voice type and make sure I’m covering all the points I want to, this becomes ninety percent of what would make me happy. And that is great. Not just “good enough”—it’s great. It is not perfect, but I would rather stick with great.