The last ten percent
Twenty-five years of building software, and it has never been easier to make something that looks finished. Or harder to make something that works. The first post in a series on software development in the age of AI and agents, from someone who builds it every day.
Open any serious platform for project work and search for one phrase: "about 90 percent done". You will find it in listing after listing. The app is almost finished, it just needs completing. The prototype works, it just needs stabilising. Someone has already done most of the work, all that is missing is someone to close the rest.
Those listings have always existed. There have never been this many. And that, to me, is the most honest picture of what is happening to software development right now. Not the demo videos, not the conference talks, but that queue of products that got to ninety percent and stopped.
What I learned about transitions
I have been building software for twenty-five years. That is long enough to stop believing in revolutions and start treating them as weather. They arrive, they pass, they leave something behind. I have been through several, and every one of them was announced, in its moment, as the end of everything we knew.
What stayed with me from all of them is one simple rule. There is only one way to get technological progress wrong: be late. Everything else can be fixed. You can misjudge a tool, miss on an architecture, invest in something that disappears. Those are mistakes that cost a month. Failing to keep up costs a career.
About twenty years ago I started working seriously for the American market, and that is where I first felt what speed means. Not the speed of writing code, but the speed of deciding. A meeting in the morning, a decision by noon, the first code that evening. Build first, figure out what to do with it later. For someone raised in a culture where every decision is thought over three times, that was liberating. Only now do I see the other side of that tempo, and I will write about it, because AI has multiplied it by ten.
The engineer nobody was hiring
All that time I carried a certain discomfort with me. I was not the engineer the industry was looking for. I cared about how a product looked, how it felt under the fingers, why someone opened it a second time or did not. The user's side interested me as much as the database did. Back then, that was not a virtue. Companies wanted narrow specialists, on a logic that sounds unbeatable: whoever does only one thing knows that thing best.
For years I had to prove that breadth is not a lack of depth. And for years the market slowly, without ever apologising, changed its mind. Today a developer is expected to understand the product, the user, the architecture, the infrastructure, the data, security, cost, and on top of all that to make decisions and stand behind them. To say that everything is expected would be an understatement.
AI did not create that expectation. It only made it merciless. While writing code was expensive, you could say there was no time to understand the product. Now that code appears faster than we can read it, that excuse is gone. What is left is one question: do you know what you are building, and why.
The revolution of bad products
Back to those ninety percent.
It has never been easier to make something that looks finished. In one afternoon you get an app with sign-in, a database, an interface that looks like a designer made it, and a landing page that promises. It has never been harder to make something that actually works. That holds under load, that can be maintained once the first euphoria has passed, that does not fall over when the thousandth user opens it instead of the tenth.
The difference between those two things is exactly that last ten percent. Projects have always broken there. What is new is that a hundred times more of them reach that point, and most of the people who brought them there do not know they have stopped. To them, the product looks done.
That is why I call this the revolution of bad products. If it used to be true that one startup in ten succeeds, I am convinced we are heading towards one in a hundred. Not because there are fewer good ideas. Because there are a hundred times more that do brilliantly in a demo.
Bad work has learned to look good
This is the part that interests me most, because nobody talks about it.
Bad work used to be visible. A bad developer shipped code that fell apart, a bad project manager a plan that did not hold, a bad analyst a document nobody could read. It was painful, but it was honest. You could see who you were dealing with.
Today a bad developer, a bad project manager, a bad analyst, whoever you like, can put in front of you, with AI, a solution that cannot be told apart from a good one. The documentation is tidy. The presentation is convincing. The code is clean, commented, well structured. It is still bad. You just no longer see it in the meeting. You see it three months later, when it costs ten times as much.
Which means that managing projects, time and decisions has not become simpler, as promised. It has become harder than ever. Because the thing intuition used to do, spotting bad work at a glance, no longer works. Bad work has learned to look good. The people who commission it, run it and pay for it have not yet learned that.
Why I am writing
With this post I am opening a series on software development from the point of view of someone who builds it every day, with AI and agents in the loop, on my own products and for clients. It will be concrete. How I work, what has genuinely helped, where a tool cost me weeks and why I let it.
But the main motive is a single one, and it is not technical. The relationship between people and AI, and the price many will pay for it in the coming years. Not because AI will replace them. Because, alongside it, they will stop doing precisely the parts of the job it cannot do for them: understanding what is being built, deciding what will not be built, answering for the result.
That price will not show up right away. It will show up in those listings. About ninety percent done, looking for someone to finish it.