Read time: 4 minutes

Welcome back, leader.

PwC just read more than a billion job ads. What they found should worry you more than any new model this month.

The first rung of the career ladder is being taken away. The job that used to let a young person get started is going to the machines.

That is not just an HR worry. It is a simple question: in five years, where do your next experts come from?

PwC looked at over a billion job ads and put out their 2026 AI Jobs Barometer.
The picture is a job market splitting into two halves.

In the first half, the good jobs are growing.
These are the roles where AI does the boring work and people get paid for judgement, for knowing what to do. They are growing twice as fast as everything else, and their pay is rising 42% quicker too.

In the second half, the door for young people is shutting.
New graduate jobs in AI-heavy fields have stopped growing. And the few junior jobs left now ask for skills you used to need years to build, like leadership. Junior roles are seven times more likely to demand those senior skills than before.

Read that again. We are asking 22-year-olds to walk in the door already senior!

Here is why this matters to you.
The boring work AI took first, the reconciliations, the first-draft numbers, the data pulls, is the exact work that used to train people. It is how a graduate slowly became a controller. PwC say it straight: AI is taking away the work that used to teach people, while asking them to be wise much sooner.

Ethan Mollick, who studies this closely, says the same thing. When AI can do good work cheaply, the thing that makes a person valuable is no longer being technically skilled. It is taste and judgement.
The catch? You only build taste by doing the boring work first. And that work is going!

So here is the hard question for any leader. If the machine does the work that used to grow your future experts, where do those experts come from?

Judgement is the rare thing now. And we are quietly getting rid of the thing that builds it.

You can't bring the old apprenticeship back.
But you can build judgement into the work on purpose, instead of leaving it to chance. That's the part still in your hands.

You can't rebuild on-the-job training by hand.
But you can put your judgement into the work itself, so a junior person can do a bigger job than their experience should allow, and learn your way of doing things while they do it.

The tool for this is a Claude Skill or a Claude Project.
Think of it as a saved set of instructions that Claude follows every single time.
You teach it your business once, instead of starting from scratch in every chat.

Put three things into it.

First, one simple rule, the one I swear by: every number must come with its source. The file, the tab, the cell. No number without proof, and no guessing. If it can't find the source, it must say so. That one rule is the difference between work you can put in front of your board and work you can't.

Second, write down what "good" looks like. The checks you do without even thinking: what counts as a big enough number to flag, what has to add up, what a good explanation actually includes. The model can't hit a target you have never written down.

Third, give it examples. Drop in two or three pieces of your best past work, so it copies your standard and not some generic one. This is the part most people skip, and it is the most powerful. In my RACEF framework, it's the "E", for Examples.

Want a quick start? The bones look like this:

Hand that to someone two years into their career and watch what they produce. They do senior-level work from day one, with your safety rails on.
And they learn your standard by working inside it.
You also stop losing all that know-how every time someone leaves, because it now lives in the work, not just in their head.

If you'd rather we build it into your team's actual workflow, with your real data and your real standards, that's exactly the kind of work we do at Prime AI Solutions

The copyright fight just found a peace deal.
OpenAI and Getty Images announced a deal this week that lets ChatGPT show Getty's licensed photo library, over 400 million images, right inside its answers. It only shows up for factual questions, like "what did the 1969 moon landing look like," not for AI-generated pictures. Getty's share price jumped over 100% the same day. For anyone who has worried about AI and copyright, this is the first sign of a real answer: pay to license the work, instead of fighting over it in court.

AI just gave 18 families an answer they'd waited years for.
Doctors at Boston Children's Hospital had run out of options for hundreds of children with rare, undiagnosed illnesses. They tried something new: feeding the genetic data into OpenAI's o3 model alongside clinical notes, then having human doctors check every result. Out of 376 unsolved cases, the model helped find 18 new diagnoses. One patient had waited fifteen years for hers.
It's a good reminder of why we do this work. Used well, with people still checking the output, AI doesn't just save time. It changes lives!