← Writing

The cognitive lever: why AI amplifies expertise but blinds the inexperienced

Björn Sennbrink ·

In the wake of the generative AI revolution, a clear trend has emerged everywhere, from viral clips on social media and open GitHub repositories to internal initiatives in business: ready-made “prompt libraries” and curated “skill collections” are on offer all around.

At first glance, the logic seems enormously appealing. If the industry’s leading experts package their methods and instructions into ready-made templates, surely anyone can download the file, press a button and deliver at the same high level?

It is a beautiful vision, but it rests on a fundamental misunderstanding of how generative AI actually works in practice.

AI is neither a coffee machine where you press a button and get a finished espresso, nor a script that does exactly the same thing every time. AI is a cognitive lever, and a lever only amplifies the force that is actually put into it.

What is the cognitive lever?

A physical lever lets a person lift a heavy rock that would otherwise be impossible to move. But for the lever to work, two things are required: a fulcrum to rest against, and an actual force applied by you.

In knowledge work, AI works in exactly the same way:

  • Deep domain knowledge and context are the fulcrum.
  • The ability to reason, evaluate and counter is the force applied.
  • The AI model is the lever itself, multiplying your reach and speed.

When a senior expert uses AI, whether for source code, advanced data analysis or strategy, there is a constant, almost invisible dialogue. The expert doesn’t just see the answer the model generates, but reads it with a trained eye. Subtle inaccuracies are caught immediately, hallucinations are seen through, and the contextual gaps the model missed are filled in as they go.

The lever lets the expert build entire applications or do weeks of analysis in a few hours:

[ Senior expert ] + [ Deep domain knowledge ] ---> ( AI as a lever ) ---> Dramatically higher quality & pace
[ Inexperienced user ] + [ Lack of context ] ---> ( AI as a lever ) ---> Illusion of competence (high risk)

The trap: when the lever becomes a blindfold

So what happens when an inexperienced user downloads the same “skill” from the web or picks up a template from a public library?

Because generative AI delivers its answers with impressive linguistic authority, a well-formed structure and a dazzling tone, what can be called an illusion of competence arises.

The user gets a text or a solution that looks perfect on the surface. But without the necessary fulcrum, their own domain knowledge, they lack the tools to answer the decisive questions:

  1. Is this actually correct, or does it just sound good?
  2. What assumptions has the model made in the background?
  3. How do you counter if the conditions or the input data change?

In the worst case, organisations and individual practitioners start mass-producing neatly packaged bad decisions. In the best case, the user is left holding an output they neither understand in depth nor can defend in front of a client.

Three common misconceptions about AI competence

To understand why investments in ready-made instruction libraries so often end up delivering little real value, three common myths need examining.

1. “A skill can be isolated from its context”

A prompt or an instruction is not executable source code in the traditional sense. It operates in a specific context, against a specific model and together with a specific person at the controls. To believe that you can take an expert’s way of thinking, paste it into a text file and expect the same result regardless of the user is to ignore the human intellectual effort.

2. “Ready-made templates replace hands-on exploration”

There is a tendency to look for shortcuts to AI maturity through standardisation and ready-made instructions. But deep knowledge isn’t built by following ready-made recipes. It is built through hands-on use, failure and continuous experimentation directly in the tools, the codebases and the workflows.

3. “AI is mainly about summarising and consuming”

Using AI to summarise documents or generate short answers is the simpler form of use. The truly transformative value only arises when AI is used for production and orchestration, when the tool works as a co-creator in complex systems and builds entirely new value.

The way forward: from ready-made recipes to genuine knowledge building

To truly capture the potential AI offers, the focus needs to shift from ready-made solutions to genuine knowledge building:

  • Invest in domain expertise: AI doesn’t replace the need for experience; it makes experience more important than ever. The more you understand your subject, the stronger the leverage.
  • Make room for your own control and continuity: Consultants, developers and creators need security and stable workspaces where their built-up context, history and customised environments can remain over time, without the risk of sudden interruptions.
  • Teach how to think with AI: It’s about understanding how models reason, where their limitations lie and how to critically review their answers, rather than just collecting a list of prefabricated prompts.

Generative AI is a fantastic lever for human knowledge. But remember: if you don’t put any force of your own into one end and lack a steady fulcrum to rest against, the lever won’t move anything at all, however advanced the template looks on paper.