The Hidden Cost of “I Can Clone This in Five Minutes”

The feed became impossible to scroll

While I was at the doctor, I was thinking about this new trend I keep seeing everywhere: “Hey, because we have Claude Code, I can do whatever I want. I can copy whatever I want. It will work. Period.” I mean, if you check my feed, it is almost impossible to keep scrolling without seeing it. This person cloned XYZ. This other person reverse engineered something. Another one rebuilt a product in one afternoon and, of course, it “worked.”

And I get it. I am not against the excitement. I am experimenting with the same stuff right now, building my own personal agent, Maria, because there is this new hype around personal agents and I could not properly test Meta’s personal-agent products. So I did what I always do when I cannot test something: I started building my own version.

Maria has been my personal assistant since mid-October. She runs on ElevenLabs, with my family Drive behind her, and she cannot write anything on her own. Every change she proposes waits for my confirmation button, because I still want to be the one making the final call.

And yes, at the beginning, it worked as expected. That is exactly where the marketing usually stops.

Then the boring part starts

The first demo is always the most dangerous part. You connect the pieces, you make the first flow work, you test one happy path, and suddenly your brain goes: “Holy shit, this is the future.” And maybe it is. But then you start using it for real, not for the screenshot, not for the post, not for the “look what I cloned” moment.

That is when the hidden cost appears. Keep testing. Keep learning. Keep improving. Keep maintaining. All the things that are not sexy, not viral, and for sure not fitting in a five-minute video.

Just to give you an idea: when I tried Claude or ChatGPT, they never gave me an error because, for instance, a websocket was interrupted. Maria did. At the beginning, she had this kind of issue, and I can tell you there are many other little things like that. Nothing dramatic, nothing impossible, but all of them are real.

The cost is not building the first version. The cost is staying there after the first version starts behaving like real software.

And this is the part people skip when they say, “I rebuilt this product in five minutes.” No, you rebuilt the surface. You rebuilt the part that is easy to show. You did not rebuild all the weird cases, the broken connections, the retries, the permissions, the “what happens when the connection drops in the middle of a sentence” type of stuff.

Builders do this to learn

I think the world, especially now with AI, will be divided more and more into builders and users. I am clearly in the builder side, for better and for worse. Builders love to build whatever they want, even when a product already exists and even when buying the software would be the rational decision.

This is the stupid part of being a builder. You see a tool, you like it, and your first thought is not “let me use it.” Your first thought is “how does this work?” Then ten minutes later you are in Claude Code, checking APIs and creating a prototype. Rebuilding is how builders learn, at least for me. For me, Maria is exactly that.

Users are different, and honestly they are often more sane. They do not care about rebuilding software or cloning products, even if technically it is possible. They want to use a product as is, with its limits, and move on with their life. And there is nothing wrong with that: most companies survive because users think like users.

The respect missing in AI marketing

What I dislike is not the builder energy. I love that. What I dislike is the marketing around it, because it is becoming too biased toward the first part of the story. “I built something.” “I cloned something.” “I replaced this huge product.” Nice. And then?

Then there is support. Then there is maintenance. Then there are users doing weird things, because users always do weird things. Then there are edge cases, integrations, documentation, uptime, migrations, data, security, pricing, onboarding, and all the invisible work that makes a product boring enough to be trusted.

I really hope that one day the AI hype becomes a bit less biased and more respectful of companies that created and maintain huge products. Because building a demo is one thing. Maintaining a product that people use every day is another sport completely.

And this is why the “you can rebuild this in five minutes” thing sounds like bullshit to me. Not because Claude Code is not powerful. It is powerful. Not because agents are fake. They are not fake. But because if you are a real builder, you know you will never say that sentence with a straight face.

You know it will cost you weekends. You know it will cost you days of testing and retesting. You know you will fix one thing and break another. You know the first version is just the invitation to the real work.

Alone from zero is still hard

Even if you have a full pool of talented agents, working alone from zero to zero is still challenging. Maybe even more now, in a weird way, because once your product is done, it has less visibility. The market sees the finished thing and immediately someone says, “Hey, I can copy and paste this in five minutes.”

Holy shit. 🤦

That sentence ignores the years behind the taste, the architecture, the customer support, the decisions not to build something, the ugly fixes, the stability work, the boring maintenance. It ignores the fact that a product is not only the interface you can see. A product is the accumulation of all the problems someone already solved so you do not have to.

Maybe AI will make us faster. I believe that. But faster does not mean free, and it does not mean effortless. The hidden cost did not disappear.

It just moved after the first “wow.”