Two American investors have a counterintuitive answer to the entry-level problem: stop treating management as a mid-career promotion and hand it to 22-year-olds on day one. Bob Morse of Strattam Capital and Michael Quoia of KSE Leadership argue in a Crunchbase News column that companies should keep hiring new graduates and assign them to supervise AI agents immediately, so they build the delegation, oversight and judgment skills people normally acquire a decade later, as their column argues. The provocations come from odd corners: a lakeside conversation they cite with a pharma head of AI who said he might never hire another junior engineer — “but who’s the next generation that is going to run this when we retire?” — and the culture of an industry where “managing analyst” was a put-down.
The column’s logic: senior people get more out of AI today not because they know more, but because they are good managers — they delegate clearly, check results, correct patiently, and occasionally move work to a better performer. AI agents now behave like knowledge workers, the investors write, so the craft of managing them can be taught young. They point to the service academies’ 200-year record of training 22-year-olds to lead, and to OpenAI’s own incident report on the Hugging Face breach, whose list of fixes never mentions training employees to manage agents — the failure mode the column says day-one manager hires would prevent.
The Data Behind the Problem—and the Dispute Behind the Data
The statistic anchoring the pitch is Stanford’s Digital Economy Lab revision, which found employment of 22- to 25-year-olds in the most AI-exposed occupations about 19 percent below where it would have been if it had kept pace with less-exposed peers; the study itself is careful to note the pattern is “descriptive, not causal”, and that adjustment runs through reduced hiring rather than separations. Whether AI is even the driver is contested: an LSE analysis of 407 million job postings found exposure to remote work predicts weaker junior hiring better than AI exposure, and when both are modelled together the apparent AI effect largely disappears, as Fortune’s summary of that study reports. The AT Pulse fact-check of the column notes the authors sit on boards of Strattam portfolio companies and that no outcome data yet exists for the “agent manager” proposal — it is an argument from analogy, not evidence.
This site has run the entry-level numbers all year, and the picture is genuinely mixed rather than simply gloomy. Gartner forecasts AI will create a net 1.7 million jobs by 2030 — 10.6 million created against 8.9 million lost — with deferrals hitting entry-level seekers hardest, our analysis of the forecast notes; the UK’s graduate data told a similarly two-sided story in our CS-grad article; and we have previously argued the real career value now builds in AI governance and oversight skills, a close cousin of what Morse and Quoia propose. The Brooks’s Law essay we covered in October made the same career prediction from the software side: the people who define, review and orchestrate agent work are the ones getting hired — our analysis is here.
What It Means in New Zealand
New Zealand’s version of this debate has a sharper edge: RBNZ governor Anna Breman told Parliament in September that AI could blunt job growth for young New Zealanders in the short term, the concern recorded in our coverage of her remarks — and the national unemployment rate now sits at a more-than-decade high of 5.7 percent, as RNZ reported. A proposal to make graduates day-one managers is, in effect, an argument that the entry-level rung should be re-anchored to supervision of AI systems rather than production of first-draft work.
The honest summary: nobody yet has outcome data showing graduates managing agents outperform graduates doing junior analysis — the column is a bet, not a finding, and it carries the authors’ commercial interest in AI-forward portfolios. But as advice to individuals, it costs little and hedges the actual measured risk. The entry-level tasks AI most plausibly absorbs are exactly the ones Stanford’s data shows employers have already slowed hiring for; the tasks around those systems — reviewing agent output, catching its confident errors, defining what “done” means — are the ones every measured trend in 2026 says are growing. A graduate who can show a portfolio of supervised AI work, with the errors caught documented, is arguing for the version of the job that is being created, not the one being deferred.
Sources: Crunchbase News, Stanford Digital Economy Lab, AT Pulse, Fortune