Anyone can learn AI skills from a tutorial. The next wave belongs to the people who know the theory well enough to catch a confident wrong answer.
AI skills (prompting, using models) commoditized the moment everyone got access. What stays scarce is AI expertise: the domain theory that lets you tell a right answer from a confidently wrong one. AI is stochastic and hallucinates in the same authoritative tone whether it’s right or wrong, so without the underlying knowledge you ship mistakes and feel good about it. When you already know the work, the model stops being an oracle and becomes a tool you can check. The advantage moves from access to understanding.
Key takeaways
- The models aren't the problem; the hidden requirement is the theory needed to use them well.
- AI is stochastic and hallucinates in a confident tone, so errors hide inside plausible sentences.
- Without domain knowledge you can't tell right from wrong output: 'confident mistakes at scale'.
- The self-taught web worked because feedback was honest (broken code visibly broke); AI removes that safety net.
- The next decade rewards professionalism and theory, the ability to catch a wrong answer before it does damage.
AI skills are not tools (until you have the theory)
AI is powerful and it fails constantly. Being stochastic is a polite way of saying it will give a confident answer that happens to be wrong, in the same tone it uses when it’s right. If you don’t understand what you’re trying to accomplish, you can’t tell the difference, and turning to a model to learn the theory first leaves you stuck, because you don’t yet know what to look for. Flip it around: when you already know the work, the model becomes a tool like a screwdriver or a spreadsheet, speeding up something you could do yourself and could check.
The web rewarded curiosity. AI won't be so generous
For two decades the internet ran on an honest deal: you could become a designer, developer or writer with a connection and stubbornness, because feedback told the truth. If your site broke, you saw it break. AI removes that net. The output always looks finished, reads well and sounds authoritative, so errors stay buried in plausible sentences you only catch if you already know enough to look. The feedback loop that taught a generation of self-starters is exactly what generative AI takes away.
The screwdriver was never the skill. Knowing what to build with it was. AI just makes that distinction expensive to ignore.
Fabio Devin
The next wave: informed professionals
The author has worked on AI since 2011 and built Quantum UX, an AI framework for user experience, in 2014, years before generative AI went mainstream. The market doesn’t need more people who can prompt; it needs people who know why the prompt works, where the model fails, and what good looks like. The advantage moves to those who hold the theory, because they’re the only ones who can use these systems without being quietly misled.
AI skills are using the tools (prompting, running models), now available to everyone. AI expertise is the domain theory that lets you judge whether the output is correct and know where the model will fail.
AI gives confident, well-written answers in the same tone whether right or wrong. Without underlying knowledge you can’t spot the errors buried in plausible sentences, so you ship mistakes unknowingly.
The web gave honest feedback: broken things visibly broke. AI output always looks finished, so the feedback loop that taught self-starters by trial and error disappears.
Know the work first. Then the model becomes a tool you can check and direct, rather than an oracle you trust blindly.
Want AI applied by people who know the theory? Talk to Dorve.
Start a project