Who trains the junior when AI does the junior work?

In February 2022 I posted a one-liner on X: a junior developer with good Google skills is a senior developer. People liked it because every developer knew it was half true. You became senior by looking things up, reading the docs, getting it wrong, and slowly learning which answer to trust.

These days, every morning around 4am, one of my agents reads the overnight AI news, writes up what mattered in a notes file, and lists a few things worth writing about. Other agents write first drafts of posts before I see them.

Research, summaries, first drafts. Four years ago that was most of a junior's week. It was also how the junior learned.

That is the whole essay. AI isn't mainly removing entry-level jobs. It's removing entry-level tasks, and those tasks were the training. So the useful question for any team using AI isn't whether to hire juniors. It's how a junior will learn the judgement that the busywork used to teach.

Is AI taking entry-level jobs?

Partly, and mostly by not hiring them rather than by firing them.

The best data so far comes from Stanford's Digital Economy Lab. In Canaries in the Coal Mine?, Erik Brynjolfsson, Bharat Chandar and Ruyu Chen use payroll data from ADP covering millions of US workers. Their August 2026 update, with data through June 2026, finds:

• No sign of widespread, economy-wide job loss.
• Employment of 22 to 25 year olds in AI-exposed occupations is 19% below where it would be had it kept pace with less-exposed peers. Experienced workers show no comparable gap.
• The gap has widened steadily since they first reported it in August 2025.
• It works mainly through fewer young people being hired, not more being let go.
• The declines are concentrated where AI substitutes for the work. Where it complements the work, employment is flat or rising.

The authors are careful to call these early descriptive indicators, not proof that AI caused the gap. PwC's 2026 Global AI Jobs Barometer, built on more than a billion job ads, sees the same shape from the hiring side. Across Canada, Singapore, the UK and the US, postings for early-career jobs (0 to 2 years of experience) have flatlined in the quarter of jobs most exposed to AI, the only quarter where that has happened. PwC also warns the chart isn't proof of cause. It adds that 49% of CEOs in its latest Global CEO Survey expect AI to reduce junior hiring over the next three years, against 12% for senior hiring.

Which entry-level tasks is AI taking over?

The ones juniors used to learn on.

ZipRecruiter's 2026 AI Worker Report, published on 26 October from a survey of more than 1,500 US workers in September, asked what people use AI for at work. The top three were research and information gathering (52.1%), summarising documents and meeting transcripts (41.6%), and drafting and editing text (40.9%). The report calls these "tasks that have traditionally taught new hires how to do the job."

The same survey shows who is getting the benefit, and who is getting the help.

More than half of entry-level workers (54.7%) are teaching themselves AI tools. The report's own summary of their position: "There are now fewer opportunities to learn how to work, and the bar for entry-level workers is rising."

The busywork was never just busywork. It was the curriculum.

What do entry-level jobs require now?

Senior skills, on day one.

PwC found that the most AI-exposed entry-level jobs are now seven times more likely to ask for skills that used to belong to senior roles than the least exposed ones. Its examples include process management, mentorship, stakeholder management and data-driven decision making. In the US, AI-exposed entry-level roles that picked up more than ten of these senior skills grew 35% in postings between 2019 and 2025. The rest fell by 10%.

Workday's October 2026 Global Workforce Report shows the bar moving inside AI skills too. Across roughly 550 employers, demand for basic AI skills like simple prompting peaked in January 2026 and then fell 25%. Demand for building AI tools and automating workflows rose 51% between September 2025 and July 2026.

The same report has the detail I keep coming back to. Over that period, job ads mentioned management and leadership skills 7% less, and training skills 13% less. Companies want juniors who arrive with senior judgement, while asking less for the people who would teach it.

How can a small team train juniors when AI does the junior work?

Stop treating the agent's output as the end of the task. Make it the start of the lesson. A setup I'd use:

A few habits that make this work:

• Keep the agent's sources visible. A summary with links lets a junior check whether it's right. A summary without links teaches them to trust whatever sounds confident.
• Write down why a draft changed. When someone edits an agent's draft, the reason (too long, a wrong claim, the wrong reader) is the lesson. Say it out loud, or a junior only sees the result.
• Put teaching time on the calendar. The hours AI saves on busywork are the obvious place for it. PwC's own advice to employers is to "redesign onboarding, mentorship, and training programmes" for junior roles that now need senior skills.
• Hand over judgement in small pieces. Which source to trust, which draft to send, which lead to chase. Each one is a small decision a junior can own and get feedback on.

Should you still hire juniors?

Yes, if you have a plan for how they'll learn.

Every senior person on your team was a junior who got to do the boring work long enough to see the patterns in it. If agents do all of that work and nobody does it by hand, you save money this year and have no one to promote in five.

In Human obsolescence I wrote that a job is a bag of verbs, and the useful question is which verbs have already moved. For entry-level jobs, research, summarising and drafting have mostly moved. Checking, deciding and explaining haven't, and those are the verbs a junior now has to learn sooner than any junior before them.

Give the agent the busywork. Then give the junior the reasons.