For most of my career, software was expensive to build.
Even a small feature had to be defined, designed, coded, tested and put into production. A team could only build so many ideas, so somebody had to decide which were worth the effort.
That constraint could be frustrating, but it was also useful.
It forced us to choose.
AI is changing that. A feature that once took weeks can sometimes be built in days or hours. Prototypes appear before the meeting about whether to build one has finished.
When building becomes cheap, production stops being the main differentiator. Everyone has similar tools and can produce another feature or variation of the same idea.
What becomes scarce is taste: knowing what deserves to exist, but also how the things that do exist should work and feel. It lives in the small decisions, the rough edges you remove and the polish that makes a product feel considered.
AI gives us more ways to build. It makes the quality of our decisions more important.
Every idea becomes a feature
I'm already seeing companies fall into a predictable trap. They have an idea, realise they can build it quickly and ship it. Then they do the same with the next one because it seems too cheap to say no.
Before long, every conversation ends with another feature. Nobody asks whether it makes the product better, what it complicates or whether it belongs at all.
Being able to build something in an afternoon isn't evidence that you should.
An idea can be technically easy and still be a bad idea. It can be useful to a handful of customers and still make the product worse for everyone else. It can improve one metric whilst quietly damaging the reason people liked the product in the first place.
Speed changes the cost of implementation. It doesn't remove the need for judgement.
Users pay for your output
The cost of a feature isn't just the time it takes to build. Every change asks users to notice it, understand it and adapt, and that cost doesn't appear in the development estimate.
Inside the company, the team celebrates shipping quickly. Outside it, somebody discovers that the product has changed underneath them again.
Products need to improve, but people also need a stable mental model of how they work. Too much change makes a dependable tool feel like an experiment.
Users don't care how quickly the team shipped. They care whether the product helps them do what they came to do. If it gets in their way, they'll go somewhere else.
We've seen this before
Twenty years ago, plenty of software products had become enormous collections of features. Every release added more menus, toolbars and settings. Nothing could be removed because somebody, somewhere, probably used it.
The result was software that could do almost anything but was unpleasant to use for the thing most people actually wanted.
Then simpler, more focused products began to win because they did less. They were easier to understand and quicker to learn. Simplicity became part of the product.
Now we risk rebuilding the same bloated software we spent years moving away from, only much faster.
AI lets us add features at a rate previous software companies could have only dreamed of. Without taste and restraint, it'll help us reach the same bad destination in record time.
Taste isn't decoration
Taste can sound vague or superficial. Colours, typography and animation can express it, but it goes much deeper than appearance.
Taste is having a clear view of what good looks like and caring enough to keep working until the product reaches it.
It decides whether a feature belongs, but it also shapes the small choices once that feature exists: how it behaves, the words it uses, the edge cases it handles and the rough edges it removes. Polish isn't one final pass. It's the result of making those decisions well.
Good taste gives a product a point of view.
It says this is who the product is for, this is the problem it solves and this is how it should feel to use. An idea can be perfectly reasonable and still not fit that point of view.
That's what makes taste difficult. There isn't always a spreadsheet that proves the answer. Sometimes you have to say no to a useful feature, and sometimes you have to keep working after it already works.
Taste is expensive
Building has become cheaper. Saying no and getting the details right haven't.
It means disappointing the customer who asked for a feature, resisting the competitor comparison and telling an enthusiastic person that their clever idea doesn't belong.
It can also mean choosing not to ship something after you've already built it.
That feels wasteful when organisations measure progress by output. A feature in production is visible. A decision that kept the product simple, or the extra care that made it feel finished, is much harder to see.
The best products often feel obvious because somebody did a great deal of difficult thinking on the user's behalf. The complexity still existed. It just wasn't handed over to the customer.
AI gives us options, not decisions
AI is extremely good at giving us options. It can produce ten designs, twenty names and several plausible implementations. It lets us explore an idea without spending weeks arguing about it in the abstract.
But generating options and choosing between them are different jobs.
AI doesn't have to live with the product. It doesn't have to support the feature in three years, explain the interface to a confused customer or watch a once-simple product collapse under the weight of hundreds of individually reasonable decisions.
We do.
The tools can show us what is possible and help us understand how it might work. They can't take responsibility for whether it belongs or whether it's good enough. Those remain human decisions.
The discipline to throw things away
Taste isn't something you're born with. It develops through paying attention.
You use products and notice where they feel clear or confused. You watch real people struggle with the thing you've made. You build enough bad ideas to recognise the pattern earlier next time.
Over time, you get better at leaving things out and making what remains better.
Writing is rarely improved by keeping every sentence. Products aren't improved by keeping every feature either.
AI should let us explore more ideas, not ship more of them. We can prototype five approaches and choose one, or build a feature, learn from it and throw it away. We can use the extra capacity to polish the important parts instead of making the product bigger.
The value isn't in how much the tools let us produce. It's in the quality of the choices we make.
Restraint is the advantage
There will be an enormous amount of software made over the next few years.
Most of it will work. Much of it will look competent. Almost all of it will have more features than anyone asked for.
The products that stand out won't be the ones that ship the most. They'll be made by people with a clear idea of what deserves their attention and the patience to get the details right.
When everybody can build, building is no longer enough.
The advantage is knowing what to build, how it should work and when it's finished. It's being willing to leave a good idea out because it would weaken a better one, then polishing what remains.
The advantage is taste.
