Maybe You Are Max Verstappen

A couple of weeks ago, I wrote “You are not Max Verstappen.” The lesson was that more capability is not always better if your skill is not ready for it. A faster car can become a liability when the driver cannot use what the machine is capable of giving him.

This article is about the other side of the same coin. Maybe you really are Max Verstappen. Maybe you have the experience, the instincts, and the skill to climb into almost anything sitting in the garage and get everything out of it.

So imagine walking into that garage and finding two new cars waiting for you. On one side is a Chevrolet Camaro SS. On the other is a Ferrari SF90 Stradale. You can drive either one, and nobody is worried about whether you can handle the Ferrari.

Which one do you take? You cannot answer that yet because I have not told you what track you are driving on.

Maybe it is a long, fast circuit where the Ferrari’s additional power and sophistication matter. Maybe it is a shorter, tighter track where those advantages matter less. Maybe you are not trying to set a lap record at all and are spending the afternoon learning braking points, experimenting with lines, or repeating the same section until you understand exactly what is happening.

The car does not change. And the driver does not change either. But the track does, and that changes which car makes sense.

I think we need to start thinking about artificial intelligence the same way. We spend enormous amounts of time asking which model is the smartest, which reasons the longest, which handles the most context, and which one sits at the top of the latest benchmark.

But “Which model is best?” is increasingly the wrong question. The better question is what task are you trying to accomplish.

If I am working through a complicated problem, comparing competing arguments, or asking a model to help me think through something difficult, I may want every bit of capability I can get. That may be the track where I want the Ferrari.

But other times I need a short document summarized, information extracted from a small group of files, a draft cleaned up, or some other relatively straightforward task performed reliably. I may be perfectly capable of using the most powerful model available, but the task may not require it.

There is also a practical reason to care about that distinction. Tokens cost money, and the most capable models often cost more to run. Taking the Ferrari out for one unnecessary lap probably does not matter much, but imagine driving an entire fleet of Ferraris around the track all day, every day.

That is essentially what a court system, law firm, or large company does when every routine task gets sent to its most expensive model. Why pay for all that Ferrari capability when the track only gives you room to use a fraction of it? You are still paying for the whole car even when the task never gives you a chance to get every bit of juice out of it.

At that point, model selection stops being a benchmark contest and becomes an architectural decision. The goal is not to deploy the most powerful model everywhere. The goal is to match enough capability to the work being performed.

For a long time, much of my message about AI has focused on matching the capability of the tool to the skill of the person using it. Start small, learn the technology, understand where it succeeds and where it fails, and move into more capable tools as your own competence grows.

But there is another side to that lesson. Eventually, you may have the skill to drive almost anything in the garage. Maturity then means knowing that the fastest car is not automatically the right car.

Maybe you are Max Verstappen. But before you grab the keys to the Ferrari, ask yourself what track you will be driving on today.