At Nokia, I learned that weak decisions rarely look weak when they are made.
They arrive with a convincing argument. Someone has thought the matter through and explained why this particular option makes sense. Later, in startups, I saw the same pattern again.
Sometimes two strong opinions collided. I often knew what would happen next: a third option would survive because neither side objected to it strongly enough. It was not necessarily the best option, but it allowed the work to continue.
There was at least some resistance in the process. People had to explain what they wanted, defend it and listen while somebody else pointed out what might go wrong.
AI changes that part of the work more than I expected.
The first result feels unusually good
I use AI every day, and I understand why people get excited about it. A prompt becomes a layout. A rough thought becomes something you can see. A request becomes working code before anyone has had time to book a meeting about it.
For many clients, that feels like getting control back. They no longer have to wait for a designer, developer or copywriter to interpret the idea first. They can make something themselves and react to it immediately.
That is useful. Waiting has never improved an idea merely by taking longer.
The trouble begins when fast production starts to feel like a good decision. The result looks finished enough to support the request that produced it. If the prompt asks for another section, AI makes another section. If the brief asks for three messages in the hero, it finds room for all three. If someone wants animation, it can provide animation.
The output may be competent. The original request may still be wrong.
A helpful answer can make the mistake harder to see
In client work, people regularly ask for a solution when the more useful question is still about the problem. They ask for a feature, a layout or new wording. Often there is a sensible reason behind the request.
But a designer may notice that the new feature weakens the main action. A developer may know that the apparently small change will make the product harder to maintain. Someone who has watched users struggle with similar interfaces may ask whether the extra choice helps them at all.
This is not expertise performing a little ceremony around production. It is part of the work.
AI knows a remarkable amount about design, code, communication and user behaviour. It can analyse information at a scale no individual person can match. Give it real user data, recurring support questions or a large set of observations, and it can help find patterns that would be difficult to spot manually.
But early product decisions often happen before that evidence exists. At that point, the work depends on what receives emphasis, what gets left out and which confident request deserves a second look.
AI usually meets a confident request with a confident answer. The person prompting it still has to recognise when resistance is needed.
That is a fairly demanding thing to ask from someone who came to AI precisely because they do not have years of experience in design, development or product work.
Cheap production creates more to maintain
A colleague once put the coding version of this problem neatly:
“Writing code faster is great. But coding was never the annoying part. Keeping the codebase intact year after year — that’s the hard part.”
The same problem appears in less technical work. Making one logo is faster than maintaining a recognisable brand while new requests, campaigns and preferences accumulate. Producing a page of copy is faster than keeping the same voice across a website, product, sales material and customer service.
AI makes it easy to add. It can produce another version before anyone has decided what was wrong with the previous one.
Over time, the cost shows up in small decisions. A page becomes fuller. A feature remains because removing it would require an argument. A sentence survives because it sounds reasonable on its own. None of these choices destroys a product. Together they make it harder to understand and harder to maintain.
This is familiar territory. Products have always accumulated things. AI just makes the accumulation much faster.
The work has moved
I do not think AI has made expertise less useful. It has made raw production less scarce.
That changes where the difficult work sits. Making a plausible version is becoming easy. Deciding what belongs in the final version is not.
The useful person in the room may now be the one who asks why the extra section is there, what the new feature does to the rest of the product, or whether the polished answer is solving the problem anyone actually has.
Sometimes the answer will be to keep what AI made. Sometimes it will be to change the request and try again. Quite often it will be to remove something that took only seconds to produce.
AI is very good at doing what we ask. That makes judgment easier to overlook, although it has not removed the need for it.
