Where The Juniors Went

Think back to your first real job.

I'll guess most of what you did that first year was boring. Data entry. Pulling reports. Drafting the first version of something a senior person tore apart. Sitting in on calls and taking notes nobody read. Proofreading. Formatting. Cleaning up a spreadsheet someone else broke.

You thought you were paying dues. You were getting trained. Every report you pulled taught you what the business measured. Every draft that got torn apart taught you what good looked like. Every call you sat in on taught you how deals actually got made, which was never the way the sales deck said.

That work is disappearing, and fast.

Challenger, Gray and Christmas reported this month that employers cited AI for 3,961 job cuts in September and 120,136 so far in 2026. That's about 21 percent of all announced cuts this year. Bloomberg reported that India's global capability centers, which employ around 2.4 million people doing back office and technical work for companies like JPMorgan, are automating entry level tasks and hiring fewer graduates. And here's the twist: weekly jobless claims just fell to 197,000, the lowest since July.

Total jobs are holding up fine. The bottom rung is the part getting sawed off. The jobs that exist are shifting toward people who already know what they're doing.

Great for this quarter. Ugly in five years, for every business, including yours. If nobody does the beginner work, nobody becomes a senior. Five years from now, where does your next operations manager come from? Your next account lead? Your next you?

This Sunday's strategy is about building a bench when the bench work is gone.

Here's the logic I hear every week, and it's not wrong on its face.

"Why would I hire a junior to do work AI does for a few dollars? I'll keep my experienced people, give them AI, and get more done with fewer hires."

For this quarter, that math works. Your senior people with good AI tools really do produce more than a senior plus a junior did two years ago.

The problem shows up later, in three ways.

Your seniors become a single point of failure. When your one experienced bookkeeper or project lead leaves, there's nobody behind them who's been learning the work. You're hiring from the outside at a premium, in a market where everyone else made the same call and nobody trained anybody.

Your seniors stop getting better. A big part of how experienced people sharpen their judgment is by explaining it to someone newer. Take away the junior and you take away the teaching, and teaching is how expertise gets examined.

You lose the people who'd have spotted what AI misses. A junior reading every client email for a year starts noticing the patterns no prompt was written for. That's how a lot of process improvements actually start. Nobody notices if nobody's reading.

So don't hire juniors to do work a model does better. Redesign what junior work is.

The old junior role was mostly production. Make the thing. The new one is mostly judgment training with AI doing the production. Three parts.

Operate. The junior runs the AI tools that do the first draft of the work. They're the one feeding call transcripts into the summary prompt, running the reconciliation, generating the first draft of the proposal. They learn the work by learning how to get the machine to do it well.

Check. Every AI output the junior operates, they review against the source before it goes anywhere. They're the first line of quality control. This is where the real learning happens, because to catch what the model got wrong, they have to understand what right looks like.

Explain. Once a week, the junior walks a senior person through three things they caught, three things they weren't sure about, and one thing they'd change about the process. Fifteen minutes. This is the apprenticeship.

Notice what this does. The junior still touches all the work a beginner used to touch. They still see every report, every draft, every client message. They're just seeing it as an inspector instead of a typist. And inspection teaches judgment faster than production ever did, because you're constantly comparing an answer against what you know.

You don't need an HR department for this. Here's the structure I've seen work in companies with five to forty people.

Step one: find the inspection points. List every place AI already produces work that a human should check before it matters. Call recaps, first draft proposals, bookkeeping categorizations, support replies, social posts, research summaries. Each one is a slot for the junior.

Step two: write the standard. For each inspection point, write down what "right" looks like in five or six bullets. "A good call recap includes every commitment with a name and date, every number mentioned, and any change from what was previously agreed." This is the most valuable document you'll create, junior or not. It's your senior people's judgment, finally written down.

Step three: run the checklist on the output. The junior's job is to hold every AI output up against the standard. Pass, fix or flag. Fixes they make themselves. Flags go to a senior.

Step four: make the flags the curriculum. Every flag is a lesson. When a senior resolves a flag, they spend two minutes explaining why. Over six months, those explanations add up to the knowledge it used to take three years of grunt work to absorb.

Step five: let the standard grow up too. When the junior catches something the standard didn't cover, the standard gets a new bullet. Your quality system gets better because someone new was looking at it.

If your team uses recorded calls, Fathom makes the inspection work much easier, because the junior can check a recap against the actual conversation in a couple of minutes instead of guessing what was said.

Here's a real version from a twelve person marketing agency that rebuilt its account coordinator role this summer.

Monday morning, the coordinator runs every client call from the previous week through the recap prompt, then checks each recap against the recording. She fixes small misses herself and flags anything that looks like a scope change.

Tuesday and Wednesday, she operates the reporting. A model drafts the monthly performance notes for each client from the ad platform exports. She checks every number against the source and rewrites any sentence that claims a cause the data doesn't support. That last part is where she learns the most, because "leads went up because of the new creative" is exactly the kind of claim a senior strategist has learned to distrust.

Thursday, she drafts first versions of social posts with AI and checks them against each client's voice guide.

Friday, fifteen minutes with the account director. Three catches, three questions, one process change.

Four months in, she owns recaps and reporting checks with no senior review. The director told me she'd learned more about how clients actually think in those four months than the previous coordinator learned in two years of building reports by hand.

Three numbers tell you whether this is working.

Catch rate. How many real problems the junior finds per week in AI output. It should rise for the first two or three months, then level off as the prompts and standards improve.

Flag rate. How often the junior escalates something they're unsure about. This should drop over time. If it stays high after six months, either the standards are unclear or the junior isn't learning, and you'll want to know which.

Time to independence. How long until the junior can own one inspection point with no senior review. In the old model this took a year or more. In this model I've seen it happen in three to four months for straightforward work like recaps and categorization.

If you want hard numbers on where everyone's time goes during this, a tracker like Rize shows how much of the junior's week is operating, checking and learning, versus getting buried in meetings.

If junior work is now inspection and explanation, you want different people.

Old junior profile: fast, accurate, tolerant of repetition.

New junior profile: curious, skeptical and articulate. Someone who reads an AI summary and asks "where did it get that?" Someone who can explain in plain words why something's off. Someone comfortable saying "I'm not sure" out loud.

Change your interview to match. Instead of a skills test on the old tasks, give candidates a real piece of AI output from your business, with two or three planted errors, plus the source material. Give them twenty minutes. Then ask them to walk you through what they found and how sure they are.

Most will find your planted errors. Hire the one who finds an error you didn't plant.

Plenty of businesses already have someone in a junior role whose tasks are being automated out from under them. The temptation is to quietly let the role shrink until it isn't one.

Try the redesign first. Show them the operate, check and explain model. Most people in junior roles would rather inspect than type. Some won't take to it, and that's information too. But you'll keep the ones who do, and they'll already know your clients, your quirks and your history. That's worth more than any new hire.

Be straight with them about why. "The part of your job a model can do is going to shrink. The part only a person can do is going to grow. Here's how we'll get you there." People handle change fine when somebody tells them the truth about it.

The owners I respect all think past next quarter. They're thinking about who'll be running the place in five years.

AI makes the next quarter easier than it's ever been. More output, fewer hires, lower costs. Take that win. But don't let it hollow out the pipeline of people who'll eventually make the calls AI can't. The owners who keep training people, in a new way, while everyone else stopped, are going to have the bench in 2030 that everyone else is trying to poach.

The grunt work's gone. You still need people learning the ropes. Redesign the role around judgment, put a standard behind it, and let the flags do the teaching.

FROM THE AI NEWSROOM

For operators who would rather build it with someone than build it alone. The full stack, installed alongside you, with the decisions made in the room instead of guessed at. Ninety seven dollars.

Join the Accelerator for $97

.....

List every place AI produces work in your business that a human should check before it matters.

Pick the one with the most volume. Write the five or six bullet standard for what "right" looks like. Get your most experienced person to review it.

Decide who checks that output today. If the answer is "nobody" or "whoever has time," that's your first inspection slot.

If you have someone in a junior role, sit down with them this week and walk through operate, check and explain. If you don't, write the interview test with planted errors so it's ready the next time you hire.

Then put a fifteen minute weekly explain session on the calendar. Cheapest bench you'll ever build.

Jordan

The AI Newsroom | Practical AI for people with a business to run.