What being good at applied AI really means

Applied AI. Forward-deployed engineer. Terms that barely existed eighteen months ago are now the talk of the town.

But what does it actually mean? Plainly, using the technology to solve problems in the real world: building things, deploying tools, trying to get value out of the models.

So picture the interview. Someone leans in and says, "How good are you at really using AI?"

The worst answer: "I'm amazing at it," said with real conviction. Not because the person is necessarily wrong, but because the confidence says a lot. A recent Aalto study found that the more AI-literate people rated themselves, the more they overestimated their AI capability.

Here, confidence was a surprisingly unreliable signal.

The honest truth is that the question is close to unanswerable. But like any good politician, I think it is best to take a bad question and answer a better one.

For me, it comes down to three things.

One. A permanent beginner

Do you continually interrogate your own workflow? Jack Dorsey says he looks for one thing above all: people who can reprogram their assumptions constantly. This technology is the most iterative thing most of us have ever worked with, and the only way to get good is to experiment.

Even on something as small as drafting an email. When the output comes back, is the antenna up? Could this be better, what could I ask differently next time to get closer to what I want?

It is the opposite of slipping into autopilot on the drive home. With AI, the relentless tinkerer never stops noticing and questioning. You can tell very quickly whether someone has this mindset or not.

Two. Something you have built

Have you turned experimentation into something real? Professional or personal, a way you have used AI to make your own life easier. It does not need to be clever - the only way to learn the tool is to experiment, fail, and see what works.

The proactive, possibly quite silly side project that actually exists. The marketer who stopped paying for a tool and built his own. The teacher who automated her marking. Will England, who runs a ten-billion-dollar hedge fund, signed off a company-wide email: "I used ChatGPT to write this, you should be using it too, and be proud of it."

"I'm good at AI" does not tell me much. "Let me show you what I've built with it" says a lot more.

Three. Steal from everyone

Are you borrowing widely enough? Everyone comes into applied AI from a different background and builds differently. Most people you talk to have made something cool you had not thought of.

Always be open to hearing what others have done; compare, adopt, and steal the good bits; collect from many people over time. There are many versions of right. If they are more advanced, pay attention. If not, they still likely do some things better than you.

More important still is that today's AI conversation is oddly homogeneous - open LinkedIn and it is the same tips, the same takes, everyone saying the same thing.

Can you think bigger than the AI bubble and look wider: how does another industry solve this kind of problem, and do their principles carry across? Being good at applied AI is really just being good at solving problems, with AI as the instrument. The edge is the range you bring, not the tool everyone already has.

How you build your AI system reflects how you think. It is an extension, or a replacement, of your existing capabilities. Looking at how others build gives you access to their mind and ways of working that you would never reach from your own perspective, which is why being open to it is high-leverage.

Know which room you are in

There is a name for the lane you are actually in: interactional expertise. Ed Fidoe of the London Interdisciplinary School describes it as the level between knowing nothing and being the academic - enough fluency to converse with the real experts and, crucially, to connect their work to the world. That is applied AI. You are not the frontier researcher pushing the models forward, and you are not the person who has never opened one. You are the one who knows enough to talk to the researcher, use what they build, and put it to work.

Someone six months deep in these tools can know vastly more than their boss and vastly less than the researcher across the table. The people I trust in AI know what they know, know what they do not, and adjust accordingly.