AI is a Tool not a Product

AI might be the most powerful tool of our time – and that’s exactly why so many people misuse it.

Countless products promise to solve all your problems through some kind of AI-powered workflow. At the same time, studies (like the one from MIT) suggest that excessive, uncritical reliance on AI can reduce our ability to think critically. On top of that, many AI tools simply fail to deliver the results they promise.

This raises an important question: how do you actually use AI the right way?

How AI Is Commonly Used

Many people use ChatGPT, Gemini & Co. to directly solve their tasks. This isn’t entirely wrong. If you need to proofread an email or rephrase a paragraph, this approach works just fine.

The problem starts when this pattern is applied to more complex problems. While LLMs are capable of solving many tasks, using them as a one-shot solution is rarely the best approach. It weakens your domain knowledge and often leads to fragile, low-quality results.

AI can do a lot – but maybe not everything at once. Especially when reliability and longevity matter.

How I use AI

For me, LLMs are not the final product. They are tools that help me build the final product. That difference matters.

Here’s how I approach it:

Understand your Problem

The most important step in any process is understanding the problem. If you don’t clearly know what you want, take the time to define it. Too often, we look for quick fixes to problems we haven’t fully understood.

This usually leads to longer development times, poor results, and higher costs. There’s a lot of FOMO around AI right now, but patience and clarity still outperform rushing to a solution.

Break down the problem

Once the problem is clear, break the solution into smaller parts. This helps you stay focused and identify which parts actually need AI—and which don’t.

I follow a simple principle: if I can reasonably solve a (sub)problem without AI, I do it.

Many tasks can be handled with straightforward automation. Automations have a key advantage: the same input produces the same output. There’s no randomness involved. This gives you consistent quality and makes debugging and improvement much easier.

Solve the Problem

This is where AI comes into play.

Use AI to help create the automations you need. Because the problem is already understood and broken down, you can use AI effectively—for example, to “vibe code” individual components.

Don’t take shortcuts here. Develop each task on its own and avoid cramming multiple responsibilities into a single prompt. Clear scope leads to better results.

After you created the automation tasks, fill in the blanks with AI solutions. Take your time on choosing the right model and prompt for your task.