AI Upskilling Starts With the People You Already Have

“Upskill or hire” sounds like a line item. It’s the first real test of whether an AI transformation survives contact with your organization.

The question isn’t new. It came up in every digital, agile, and data transformation before this one. What’s different now is the speed of change, and how much more of the work AI disrupts than any of those did. People haven’t changed, though. We’re wired much as we were on the savannah, so moving people still takes what it always took.

Here’s the pattern. A leader decides the operating model has to change: less project delivery, more product. There’s a north star, an all-hands, a two-hundred-person town hall. Then comes the move that feels boldest: clear the decks, bring in fresh people with the new skills, stand up a unit that works “the new way,” walled off from the old process. A startup inside the company.

It looks like conviction. It usually ends the same way, because the hard part was never the technology. The technology turns over every few months. What you’re actually trying to move is people, and people don’t arrive as blank slates. They come with a baseline: years of knowing how the work really gets done, where a process quietly depends on two other teams, which rule nobody likes but everybody needs.

“Upskill” has “up” in it for a reason. You build on that baseline.

Forget it and the new team can’t join the as-is to the to-be; the old world and the new world run in parallel and never meet. It’s the most expensive mistake in this business, and companies make it at any speed. It happens whenever a plan covers only the new skills and leaves out the ones the work already runs on.

Most leaders carry a bias without noticing it: whatever comes from outside beats whatever’s already inside. It’s comfortable, because it lets you skip the slow work of developing the people in front of you.

But there’s no outside to raid. Ten years ago we had digital natives, a generation that arrived fluent. There’s no equivalent now. AI is new for everyone, so the market you want to hire from is about as thin as the bench you’re trying to replace.

That changes what you screen for. A named tool on a CV is a snapshot of a moving target; whatever’s current today is dated in two quarters. What lasts is learning agility: the capacity to pick up the next thing, and the one after. Treat AI literacy as a proxy for it. A university degree today covers the bare minimum, so anyone fluent now went and built the skill on their own, and that tells you how they’ll handle the next shift.

It’s also why you keep hiring juniors. Hire only finished, senior skills and you build an organization of mercenaries who close one assignment and move on. That works if your whole deal with people is transactional. If you want a strong culture, you need people who move from team to team and end up knowing several sides of the business, with people they trust in each. That takes career paths, and career paths start junior. For the avoidance of doubt: junior in the AI era says nothing about age.

The learning-agility screen. Four signals to read in an interview, each with a question to ask and what a strong answer sounds like.

Done properly, upskilling is a blend, and you can run it three ways. All three keep the baseline in the room.

The fourth pattern is the one that keeps failing: new skills in their own team, rebuilding the process from scratch, parallel to the legacy team still doing the real work. On a slide it’s clean. In practice it dies, cut off from the day-to-day knowledge that keeps the thing running. I’ve watched it in something as unglamorous as invoice admin, and in a recruitment pipeline: the fresh team automates the obvious path, hits the legacy (the exceptions, the dependencies, the reason an awkward rule exists) and stalls. Nobody in the room has lived the as-is, so they can’t bridge it to the to-be.

The join is where transformations hold or break. So test for it before you commit budget.

A real competence sits on three layers, and most programs build only the top one.

The part budget-holders miss: skills don’t pay off as an event. They pay off inside a transformation: new ways of working around them, clear expectations, the right tools, and incentives that reward the new behavior over the old. Part of that is saying the quiet thing aloud. People who want to keep working the old way can, just not here. The same holds for hires. Drop a new skill into a system that hasn’t changed around it and it stalls too; “just bring in the skills and the rest takes care of itself” is how expensive hires quietly go nowhere.

For whoever owns learning, the lesson is blunt. Start by building the learning agility of the people you already have, and design the program around all three layers. A catalog of courses only covers the “being able” layer. The harder job (the knowing, the wanting, and the change around it) is what decides whether any of it pays back.

You can’t move everyone the same way, and it has nothing to do with age. When people meet a change, they fall into four groups. Learn to recognize them in a meeting, because each one needs a different move.

Don’t sort these by birth year. Plenty of people near the end of a career won’t budge; plenty in their fifties have led three transformations and are exactly who you bet on. A good share of the biggest companies around were built by people in their fifties. You’re reading willingness and learning agility, not the decade someone was born in.

At work, and especially in a transformation, the brain is constantly scanning for threats and rewards. Whichever group people fall into, you want them to take up the change instead of reacting with fight-or-flight. The SCARF model gives you five angles to keep in mind when you design the change approach.

Every technology shift runs the same sequence. Most of the pain comes from forcing a later phase too early. Before you push, work out which phase you’re in.

It starts as an invitation, and nobody has a target yet. You let the early adopters try things, build the muscle, and find where the value actually sits. Then it formalizes: now that the value is visible, you weigh it against the effort, set real objectives, and put money behind the places that pay back. Then the door closes. The value is proven, and it’s adopt-it-or-move-on.

Email went exactly this way: early adopters first, lawyers still faxing years later, then one day it was simply how everyone worked. Your pace is set by competitive pressure, not preference. Advertising ran all three phases in months because the value was obvious and the downside small; most industries get longer to breathe.

The failure mode is easy to name, and it’s the point of locating yourself on the line: hard adoption targets while most people are still being invited. Change already triggers a threat response (ordinary human wiring) and a mandate dropped too early turns that into fight-or-flight. If you’ve set targets but your people are still in the invite phase, you’ve skipped a step, and the resistance you’re fighting is one you created.

The costliest mistake in transformation is also the most repeated: throw out what you have, start clean with something new, and let the old and new worlds drift until they never touch.

One question keeps you out of it. Ask where are we scaling up from? and you’ll upskill what you have instead of discarding it. It doesn’t have to be the company’s whole operating model; it can be a single process. You’ll find the baseline worth building on, and the people worth building it with. Skip it and you’ll keep funding clean-slate teams that look fast and quietly come apart at the join.

So ask it first. Before the job descriptions, before the platform, before the town hall. Everything that holds starts there.