28,000 jobs a month, none of them cut: the bottom rung is gone

28,000 jobs a month: that is what US tech and finance payrolls are shedding in 2026, on government data (Bloomberg, July 1, 2026). No layoff waves. No town halls. No severance headlines.

The jobs are not being cut. They are not being created.

Goldman Sachs reads the same numbers and calls it hiring suppression: firms using AI to avoid adding headcount rather than to remove it. The displacement is real, but it is invisible. Nobody films a job that was never posted.

And it does not land evenly. It lands on the people trying to get in.

The last field note argued that AI took the easy part of the work and sent the bill for the hard part to whoever still knows how to do it. This one is about who gets that bill first: the people who were never given the chance to learn the hard part.

The entry-level job did not disappear, it got promoted

PwC scanned more than a billion job adverts across 27 countries for its 2026 Global AI Jobs Barometer. Buried in it is the most important hiring statistic of the year.

AI-exposed entry-level roles are now seven times more likely to require what used to be senior-level skills: judgment, leadership, adaptability. Those roles have grown 35% since 2019. Ordinary entry-level roles, the ones you could walk into and learn, shrank 10%.

Sit with that for a second.

The first draft, the data pull, the ticket triage, the meeting notes. That was never busywork. That was the apprenticeship. It is how every senior person you know became senior: thousands of small, low-stakes reps, taken under supervision, until the pattern-matching became judgment.

Those reps are now the machine’s reps. What is left for a first job is the part you used to need five years to be trusted with.

While the bar rises, the proof collapses

Here is the cruel twist. At the exact moment employers need stronger evidence of judgment, every traditional signal of it has stopped carrying information.

Greenhouse’s 2026 AI in Hiring Report puts numbers on it: 91% of recruiters have run into candidate deception. 65% of hiring managers have caught applicants using AI deceptively, reading from generated scripts (32%), hiding prompt injections inside CVs (22%), or turning up to video interviews as deepfakes (18%). 74% say they are more worried about fake credentials than they were a year ago.

The CV, the reference, the polished interview answer: all three were proxies for competence. AI can now fabricate all three at near-zero cost, so all three are worth near zero.

Both sides are automating past each other. Candidates use AI to look competent. Employers use AI to filter the flood. The flood is winning, and the honest junior with real ability is indistinguishable from the confident one with a good prompt.

What employers are actually buying now

Harvard Business Review published the demand side on July 8 (Doucette and Gaur, Cornell): a study of the three sectors that recruit the most MBA graduates. The finding is not that AI replaces expertise. It raises the bar for it. The profile that gets hired pairs domain depth with AI fluency, critical judgment and systems thinking. On day one.

Put the four reports together and the repricing is clear.

→ Fewer doors: hiring suppressed, not slashed, so nobody protests. → Higher bars: the surviving entry roles demand seniority up front. → Broken signals: no reliable way to prove you clear the bar.

Companies used to buy potential cheap and season it into judgment on the job. Now they want the judgment delivered pre-seasoned, they have automated the seasoning, and they cannot verify anyone’s claims.

That is not a tough labour market. That is a market failing the routing.

What you do with this

If you hire, two moves.

Rebuild the apprenticeship on purpose. PwC’s data says juniors with ready-made senior judgment barely exist, so if you will not grow them you will bid against everyone else for the same few profiles. Give juniors the reps the machine took, deliberately: reviewed drafts, shadowed decisions, small owned calls with a safety net. It now costs money it used to cost nothing. Budget it like the infrastructure it is.

Verify work, not claims. If 91% of your peers are seeing deception, the CV screen is theatre. Move the funnel onto demonstrated judgment: real work samples, short paid trials, a conversation about an actual decision the candidate made and what it cost. Slower per candidate, faster per hire.

If you are early in your career, stop polishing the CV. It is a broken signal on both ends of the pipe. Build a public body of judgment instead: ship something real, however small, and write down the decisions you made and why. A document can be faked in a prompt. A track record, visible and dated, cannot.

The juniors were not automated out of work.

The work that made juniors into seniors was. Rebuild that, or the shortage in five years will not be at entry level. It will be everywhere above it.

We didn’t automate the juniors. We automated their apprenticeship.