AI can write the code. It can't decide what to build.

There is a story going around that software is now easy, because AI can write the code. Half of that is true. AI can write code, quickly and often well. The trouble is the quiet assumption underneath it, which is that writing code was the hard part. It never was.

The easy part and the hard part

Writing code is the visible part of building software, so people assume it is the whole job. It is a bit like assuming the hard part of building a house is laying bricks. Laying bricks is skilled and necessary, but the decisions that make or break the house happen before and around it: where it goes, how it holds up, what happens when it rains.

Software is the same. The typing is now largely handled. The judgement is not, and judgement is where the value lives.

The decisions AI does not make for you

Ask an AI to build a feature and it will build a feature. What it will not do is stop and ask whether you should build that feature at all. It does not know that your real customers behave differently from the obvious assumption. It does not know that the simple version would create nine tenths of the value at a fifth of the cost. It does not know which corner is safe to cut and which one will bring the whole thing down in three months.

Those are judgement calls, and they come from understanding a business and understanding software at the same time. AI has neither. It has fluency, which is a different and more dangerous thing, because fluent and wrong looks exactly like fluent and right until it matters.

Why "it doesn't know what it doesn't know" is the whole problem

An experienced builder knows where the traps are, so they check for them. AI does not know a trap exists, so it walks straight past it with total confidence, and produces something that looks finished and works in the demo. The gap only shows up later, under real load, with real data, in front of real users, which is the most expensive place for a gap to show up.

This is not an argument against using AI. We use it every day, and it makes good people faster. It is an argument against the fantasy that the tool removes the need to understand what you are building. If anything it raises the stakes, because it lets you get much further down the wrong road before you notice.

What this means for you

If you are thinking about building something, the reassuring news is that the tools really have got dramatically better, and real software really can be built far faster than it used to be. The important news is that speed is only safe in the hands of someone who understands both your business and what is happening under the bonnet.

AI writing the code is genuinely a revolution. It just revolutionised the easy part. The hard part, deciding what to build and knowing when something is quietly wrong, is exactly as human as it ever was.