AI Is a Car, Not the Destination

I keep coming back to one analogy for AI. AI is like a car.

Not a magic carpet. Not a sentient assistant. Not a replacement for your judgement. A car. A tool that makes certain journeys faster and easier, that extends your reach and reduces friction, and that you still have to learn to drive.

What the car does well

A car covers ground that would take hours on foot. It carries weight you couldn't lift. It lets you be consistent when your legs are tired. In the same way, AI handles volume that would exhaust a human, finds patterns across data that no person would have the time to spot, and runs processes at a speed and scale that makes previously impossible workflows practical.

That is genuinely valuable. Not hype. Not science fiction. Real, usable productivity. The organisations I have worked with that get this right treat AI as a vehicle: they ask where they are trying to go, and whether AI makes that journey faster or more reliable. When the answer is yes, they use it.

What the human still owns

The car does not decide where you are going. You do. You choose the destination, pick the route, decide when to take the motorway and when to go cross-country. You watch the road conditions, brake for hazards, and override the satnav when it sends you down a dirt track.

That is not a limitation of the car. That is the appropriate division of responsibility. The car has no stake in the destination. It has no judgement about whether you should be making this trip at all. It does not know you are running late, that your passenger has a bad back, or that the road you chose was the wrong one three miles ago.

AI is the same. It will optimise toward whatever objective you give it, follow the path you set, and execute at scale. The judgement about what to optimise for, and why, and at what cost, remains entirely yours.

Autopilot does not mean hands off

Modern cars have autopilot features. They keep you in lane, adjust speed with the traffic, take some of the strain from long stretches. They are genuinely useful.

But experienced drivers know the rule: autopilot assists, it does not replace attention. The moment you stop watching is the moment you are unprepared for the thing the system did not expect. Edge cases. Unusual situations. Circumstances the model was not trained on.

AI assistants and agents work the same way. The failures I have seen in production AI systems are almost never model failures. They are failures of oversight: moments where someone decided the system was reliable enough that close attention could be relaxed. It almost never can be.

The stairs test is the wrong test

A car cannot climb stairs. If that were your test for whether cars are useful, you would conclude they are not. That is a strange conclusion to reach.

I hear a version of this applied to AI regularly. "It still cannot do X." Where X is some task that requires continuous embodied experience, genuine empathy, or common sense built from decades of physical and social reality.

The car was never designed to climb stairs. AI was not designed to replace everything humans do. That is not the right evaluation. The right question is: where does AI extend human ability, reduce friction, and make previously slow or expensive journeys possible? In those places, it is worth applying. In the others, use your legs.

Knowing how to drive

The future belongs to people who know how to drive AI well. Not the people who delegate everything to the model and check out. Not the people who dismiss it as a toy. The ones who treat it as a capable vehicle: they understand its limits, keep their hands close to the wheel, and know when to take back control.

For leaders, that means building teams with that judgement, not just teams with access to tools. It means designing workflows where AI does what AI is good at and humans do what humans are good at. It means being honest about where the model's confidence ends and where yours has to begin.

The organisations that assume the car will figure out the destination are the ones that end up in the ditch. The ones that thrive are the ones where a human is still, clearly, in the driving seat.