Software has always moved with its users. Long before AI, the interface was already being rebuilt around new generations. Before, we were looking for the content. Today, the algorithms made the content comes to us. They succeeded to read us, better than we do. Predicting what we want before we even think about it. Prediction engines running on years of accumulated behaviour are now our daily experience with technology.
i’m not really a social media guy, but the day i discovered TikTok, i immediately thought that this was the best content platform ever created. The algorithm was perfect to me. It always succeeded to suggest the exact content type i needed. that i probably wanted inside. In just a few hours of scrolling, it perfectly identity my taste and what i will like the most. TiktTok just did it well, better that others but it …
Everyone is a wide open book for machines.
Minimalism is beautiful, and it is not the antidote
Apple was for a long time the absolute reference to tasteful (i think they still are). Every product was well crafted, from the hardware to the software. Their ergonomy and ecosystem was unmatched and felt almost magic sometimes (nf. airpods, experience with multiple apple devices, etc.).
it clearly was the inspiration for almost every company. Even the biggest competitors was forced to copy them in the UI.
Today, i have the feeling that we have a wave of interface inspired by Apple. And it’s the cleanest we have had in twenty years. The philosophy of Apple is now a standard :
A feed with immediate content. A single input. Three buttons, four at most, and everything secondary tucked away behind a settings page.
A frictionless interface is not simply a gift to the user. Every decision you remove is also a moment of awareness you remove, a moment where someone might have stopped and asked why they were there at all. The feed and the three buttons are not accidentally addictive. They are addictive because they are effortless. Minimalism is not the cure for the attention economy. It is its delivery mechanism.
simplicity means few things, all of them reachable. Opacity means few things, and the rest sealed. Most companies sell you the first and ship you the second.
Intelligence is the product
Every tool is becoming agentic. Every product is learning to talk to other products. It is messy, half of it is duct tape, and it is still the most interesting thing to happen to software in a decade. We are at the very beginning.
I am convinced that software is about to become genuinely personal : really “personalised”. Shaped to how one specific person actually works. Consider what a model can do with your habits, your history, your vocabulary, the shape of the problems you keep returning to. That is not a feature. That is an architecture. AI will sit at the centre of it, in professional tools and personal ones alike.
Which is why the Ask AI button is a transitional artifact. It is a menu entry for something that should be a property of the whole system.
Intelligence should be ambient: present everywhere, aware of context, arriving when it is needed and silent when it is not.
A button still matters, Apple kept a way to call Siri, but the best button is the one you rarely need.
And this is the philosophy i’m trying to build in Alabasta : a project system should notice that the ticket (issue, task, or whatever) you are writing is a duplicate of one filed three weeks ago, because it can read both. It should be able to tell you who on the team is the right person for a piece of work, because it has seen who solved the last four problems that looked like it.
No prompt. No modal. No button.
But that ambition creates a problem I had to think hard about, because it collides with something else I believe.
Intelligence must assist you, not owe you
If intelligence is perfectly invisible, you never notice the handoff. You do not experience delegating a decision. You simply stop having the thought, and nothing tells you that you stopped. You lose your cognitive advantage as human.
That is the difference between autocomplete and autocorrect :
- Autocomplete proposes and you accept. You are still the author.
- Autocorrect decides, and you find out later, sometimes years later, that you no longer know how the word is spelled.
So the goal is not visible AI, and it is not invisible AI. It is legible AI. The rule I now design against:
Ambient intelligence may be invisible in its effort. It must be visible in its effect. Every action it takes should be attributable, refusable, and reversible.
We ask and we execute
We have never been this productive. The work is faster and often cleaner. That part is real, and I have no patience for people who pretend otherwise.
The cost is somewhere else. We do not write code the way we used to. We do not sit with a design. We do not turn an architecture over in our heads for two days before touching a keyboard. We ask, and we execute. And underneath it there is a quiet assumption doing enormous damage: it knows better than I do.
I used to answer this by saying that machines beat us technically but never on creativity. I have dropped that argument. We spent years insisting a machine could not produce good creative work, and it did. We were certain it would not touch software jobs, and it did. Betting again on the same square, with the same confidence, is not a position but a habit.
What has to stay ours is narrower and more durable: choosing which problem is worth solving, and being answerable for the result. Taste and accountability. A model can imitate understanding well enough to fool most reviewers, and it may eventually predict you better than you predict yourself (TikTok). But when a decision harms someone, there has to be a person who chose. Understanding is a capability question, and capability keeps moving. Responsibility is not a capability question, and it does not move.
Protecting that is not about arrogance. Believing you are smarter than the model is not an escape route; it is how you get replaced with confidence. What protects you is a practice. Keep solving things unaided, on purpose, regularly. Explore the idea in your own head first. Find the limits yourself, sketch your own solutions, and then bring the model in to attack them.