A junior product manager shipped a working landing page in 14 minutes last month using nothing but prompts. The page was beautiful. It converted at half the rate of the page it replaced. Nobody on the team could tell her why.

That is the story of 2026. Output is now cheap. Noticing is now the premium.

What AI actually changed

Figma’s 2025 state-of-AI-in-design data captured the gap in one number. 78 percent of designers say AI makes them more efficient. Only 47 percent say it makes them better at their role. The split is the whole point. Speed went up. Judgement did not.

The same thing is happening across engineering, data science, product, writing. Everyone can now produce output. Very few people can tell you if the output is any good.

The art of noticing

Noticing is a trained skill. It looks like this:

• Seeing what is off. A generated UI with eight pixels of unexplained whitespace. A model that is 92 percent accurate on the wrong metric. A paragraph that is grammatically perfect and emotionally flat.

• Naming why. The next step after noticing is being able to say the reason out loud, so the next version does not make the same mistake.

• Killing good-enough work. The designer Natalia Akhmedov wrote this year that most AI-era portfolios are “technically flawless and strategically empty”. Nobody weighed two options. Nobody chose.

• Serving a specific user, not a generic one. AI generates competent averages. Noticing is what pushes a product toward one real person.

That last one is the whole gap. Nielsen Norman Group defines taste as the ability to “orchestrate hundreds of small decisions around a central vision”. AI generates four reasonable layouts. Only a trained human knows which one serves the user.

Why this is good news

Noticing is not a talent you are born with. It is a skill you build by shipping real work, getting feedback, and iterating in public.

At Amsterdam Tech

Our Bachelor in Software Engineering and our Professional Masters in Data Science and Leadership both run on this logic. Students build real projects for real users and defend their choices in critique. By graduation, they have the one thing AI is still bad at. Judgement.

What is the last piece of AI output you looked at and knew was off but could not quite say why? Our Software Engineering brochure walks through how we train that muscle.