AI in Your Career

Familiar isn't fluent

Almost everyone has used AI by now. Far fewer have become fluent with it, and the gap between those two states is where the professional advantage sits.

Surface familiarity looks like: you open a chat window when you're stuck, you accept or reject what comes back, and your results swing wildly depending on the day. Fluency looks like: you know before you start whether a task suits AI, roughly what good output will look like, and what to do when it isn't that.

Fluency isn't knowing more tools. It's having reliable judgement about when to reach for one, and knowing what to do when the first attempt disappoints.

Six steps of this path were about building exactly that judgement. Here's an honest look at where yours sits — each row links back to where it's covered.

Finding the leverage in your own role

The highest-value AI use in your job is rarely the one in the demos. It's whatever you personally do that's repetitive, verifiable, and currently eating hours — and that's different for a lawyer, a nurse manager, a data analyst and a backend engineer.

Rather than a list of use cases that won't match your job, here's the test that generates them. Tick what's true of a task you actually do:

The pattern behind the scoring: verifiability and volume create leverage; private context and undetectable risk destroy it. Tasks in the first group are where you should start, and they're usually unglamorous — the reformatting, the first drafts, the summarising, the boilerplate.

Staying current without drowning

The field produces more news per week than anyone can absorb, and most of it won't matter to you. Trying to track everything is the fastest route to burning out and quietly disengaging.

The filter that works: ignore capability announcements, track things that change how you'd build. A benchmark score moving is news; a new way to give models access to your tools changes your options. Most weeks, nothing qualifies — and that's fine.

The other half is that your own knowledge goes stale silently. Things that were true a year ago — models can't do long documents, they can't use tools, they can't see images — quietly stopped being true, and people are still designing around limits that no longer exist. Re-test your assumptions occasionally rather than inheriting them from your past self.

Talking about it credibly

"Proficient with AI tools" on a CV means nothing now — everyone writes it, and it survives no follow-up question. What lands is specific and outcome-shaped:

• Name the problem, then the result. "Our support team spent a day a week tagging tickets; I built a classifier that does the first pass, and they check it in an hour."
• Be precise about the division of labour. Saying what you kept for yourself — judgement, verification, the final call — signals maturity far more than claiming AI did everything.
• Mention what didn't work. Anyone who has genuinely built something has a story about where the model failed and what they changed. That story is the credential.

The same applies inside a team. The person who's trusted on this isn't the loudest advocate — it's the one who can say "this won't work here, and here's why" as readily as they say "let's try it."

The part that's yours

Using AI at work shifts what you're responsible for, but never how much. Four things stay yours regardless of what produced the output:

• Accuracy. "The AI wrote it" has never once worked as an explanation for a wrong number reaching a customer. You signed it; you own it.
• Confidentiality. The paste test from Step 03, every time — and know what your organisation actually permits rather than assuming.
• Attribution. Norms differ by team and context, but being quietly evasive about how something was made is a bad position to be in when someone asks.
• Judgement about people. Anything touching hiring, performance, or someone's access to something deserves a human decision — models reproduce patterns in their training data, and those patterns include historical bias.

Make it compound

The people who get dramatically better at this don't study more — they build a small number of habits that accumulate. Pick the ones you'd actually keep:

The last one is worth more than it looks. Explaining something forces you to find the edges of your own understanding, and it's the fastest way to discover which parts you only thought you knew.

You've finished the path

Seven steps ago the goal was to stop being someone who occasionally types into a chat box. You now know what a model is and isn't, how to ask it for things properly, which tool suits which work, how to call one from your own code, how to give it your documents, and how to let it act with appropriate supervision.

That's a genuinely uncommon combination — not because any single piece is hard, but because most people stop at step one or two. The rest is repetition on real problems.