AI Took the Grunt Work—So Juniors Lost Their Classroom

Why It Still Starts With a Pencil

Pininfarina, the Italian design studio behind Ferrari and Peugeot, has one rule for its junior designers: when conceiving a new car, you must start with a pencil. In an age when AI can spit out 200 renderings, why insist on a hand sketch?

This week’s Financial Times looked at how AI is reshaping the way junior employees work and learn. Rather than simply reviving the old model of hiring fresh graduates, global companies are now redesigning the very path by which juniors learn the job. It’s worth watching in Korea too, where “junggo-sinip” hiring—bringing in juniors who already have some work experience elsewhere, rather than raw graduates—has been on the rise.

Entry-Level Hiring Fell, But Outlook Rebounded

Let’s start with a 2023 study. A Harvard Business School research team ran an experiment on 758 Boston Consulting Group consultants: the group using GPT-4 completed 18 consulting tasks 12.2% more thoroughly and 25.1% faster. But on complex management tasks that fell outside AI’s competence, the group was actually 19 percentage points less likely to reach the right answer. The research team called this uneven boundary the “jagged frontier.” Some of the research, analysis, and writing work that AI can now handle used to be exactly what entry-level hires did.

Employment data shows a similar shift. According to the “Canaries in the Coal Mine” paper published by Erik Brynjolfsson’s team at the Stanford Digital Economy Lab, early analysis found that after generative AI’s spread, employment among 22-25 year-olds in occupations with high AI exposure fell by roughly 13% relative to other groups — while employment among experienced workers in the same occupations stayed stable. In Brynjolfsson’s words, “what younger workers know overlaps heavily with what LLMs can replace.”

This year, some companies’ hiring plans point in a different direction. In the National Association of Colleges and Employers’ (NACE) Spring 2026 hiring outlook, employers said they plan to increase new college graduate hiring by 5.6% this year. That’s a reversal from just six months earlier, when the same survey put the figure at -2.4%. It’s worth noting this measures hiring plans, not actual outcomes — the survey covers only 185 companies, and the increase is concentrated among large employers with 5,000 or more staff, who reported 8.7% growth.

More interesting is an analysis linking spending and workforce data across roughly 21,000 U.S. companies, conducted jointly by Ramp and Revelio Labs. Companies that invested most aggressively in AI increased overall hiring by about 10% over the two years following adoption — and entry-level hiring by 12%. In other words, the companies using AI the most are also hiring the most new graduates. This isn’t proof that AI creates jobs, of course. A more natural reading is that companies successful enough to invest heavily in AI also happen to be hiring a lot of people generally — and the research team attached exactly that caveat.

LinkedIn’s Aneesh Raman sums up the shift this way: “We’re starting to see signals that entry-level work is moving from grunt work to real work.” The old kind of entry-level job isn’t coming back as-is — what’s changing is the substance of what entry-level hires actually do.

Grunt Work Used to Teach the Job

There’s something worth pausing on here. The tasks we used to call grunt work — cleaning up meeting notes, drafting research, reviewing contracts, revising drawings — were never just busywork. They were also a chance to absorb an organization’s tacit knowledge¹. Doing the grunt work alongside a senior colleague, learning “why this organization judges things the way it does” — that was the core of the apprenticeship model.

Now that AI handles much of this work, organizations that haven’t built a separate training track risk cutting off that learning opportunity for newcomers. The problem is that companies have shrunk the space for learning on the job while simultaneously demanding seasoned competence from day one. PwC’s 2026 AI Jobs Barometer, which analyzed over 1 billion job postings across 27 countries, found that entry-level postings demanding senior-level skills — “seniorized” entry roles — have grown 35% since 2019, while entry-level postings without such demands have fallen 10%. In occupations with high AI exposure, 52% of the skills newly appearing in entry-level postings were skills that used to be required only of experienced hires. In occupations with low AI exposure, that figure was just 7%. That said, not all of the change since 2019 can be attributed to generative AI alone.

Companies are cutting the chances to learn while demanding skills that are already fully formed. This gap shows up in education, too. A Pearson-AWS AI readiness report surveying 2,700 people across 6 countries found that 78% of university leaders believe their graduates meet employers’ expectations — yet 53% of employers say they struggle to find graduates with adequate AI skills. Only 14% of graduates rated their own practical AI proficiency as high.

NYU’s Gary Marcus adds a further concern: newcomers who grow up leaning on AI never develop critical thinking, and as a result, never develop the ability to catch AI’s mistakes either.

Pininfarina’s pencil rule resurfaces here. Daniel Lee, chief designer at the Shanghai studio, explains the problem with AI renderings this way: in the past, if the proportions were off, it showed up immediately in the lines. Now, flashy colors and backgrounds paper over the mistakes. That’s how you end up with renderings of internal-combustion cars missing the radiator vents entirely — a basic-of-basics error. Because proportional mistakes are much harder to hide when drawing by hand, it’s used as a training exercise to help newcomers master the fundamentals.

Three Ways to Rebuild the Onboarding Path for New Hires

The responses from leading companies fall into roughly three categories.

First, using AI to support new hires’ learning. Germany’s DHL trains AI on official manuals while having veteran employees nearing retirement supplement and correct that knowledge. It’s a project that turns tacit knowledge—which might otherwise vanish—into a shared organizational asset. The consulting firm Cognizant built an AI called “Harness” to guide new employees, which CEO Ravi Kumar compares to a self-driving system built on the accumulated experience of millions of drivers. It’s an attempt to shift some of the knowledge that used to come from asking a colleague directly into a system instead.

Second, redefining the role of the middle manager. Kumar believes middle managers—who once measured and coordinated—need to become “player coaches” who directly develop new hires. The CEO of legal AI company Luminance offers a more radical forecast: only juniors who understand AI and seniors whose productivity has been boosted by AI will remain, while the middle layer disappears entirely. Both outlooks anticipate a shift in the middle manager’s role, though the actual change will likely vary by organization and function.

Third, restoring in-person collaboration. The engineering consultancy Arup uses a learning model built on 70% hands-on work, 20% learning from colleagues, and 10% formal training. This shouldn’t be read as a precise measurement of what actually drives growth. Niels Fischer of Zaha Hadid Architects goes a step further, suggesting that it may have been remote work, more than AI, that eroded junior employees’ soft skills. These cases show that alongside adopting new tools, companies also need to design environments where people learn together with colleagues.

Korea Already Cut Entry-Level Training Before AI Arrived

Comparing this global trend to Korea’s situation reveals a problem. Korea is a market that had already been shrinking the number of positions meant to develop newcomers, even before AI became widespread.

It has been several years since large conglomerates shifted from scheduled mass hiring to rolling recruitment, and the seats that opened up have gone to “junggo-sinip” (literally “used-but-new” hires), meaning entry-level candidates who already carry some work experience. According to an Incruit survey, 28.9% of last year’s new university-graduate hires were these experienced newcomers — up 3.2 percentage points from 25.7% in 2023. HR managers said the cutoff for what still counts as “entry-level” experience is 3.1 years. And among job seekers who haven’t yet landed their first job, 73.8% said they intend to apply as one of these experienced newcomers going forward.

Unlike the overseas cases discussed above, hiring practices built around these experienced newcomers let companies offload training costs onto job seekers and other firms. The more widespread this practice becomes, the harder it gets for people with no experience to land a first job — and the smaller the eventual pool of skilled workers available for companies to hire later. If AI takes over entry-level tasks, companies will need to design more concrete learning paths for newcomers.

So I read this FT piece as an occasion to reconsider how Korean companies train new hires. AI-assisted training, mentoring from senior colleagues, and in-person collaboration are all methods worth testing depending on an organization’s needs. Whichever approach is chosen, it has to guarantee newcomers real chances to make judgment calls and receive feedback.

Oswarld’s Lens

I consult on GTM strategy and organizational design, so I sit in on hiring discussions often. What always strikes me is how easily a room can decide “let’s not hire entry-level for a while.” That decision, in effect, means giving up on growing your organization’s future seniors internally and instead buying them as experienced hires from the outside market. When this decision repeats across many companies, it can erode the foundation the entire industry relies on to develop new talent.

The line I found most important in this piece is from LinkedIn’s Raman: “the work of entry-level employees offers a preview of where all work is headed.” What’s happening to entry-level workers right now — routine tasks shifting to AI while humans move toward judgment, verification, and coordination — could show up at every other level too. The company that solves the entry-level onboarding problem is, in effect, the company that first solves the organizational design problem of the AI era.

So here’s my one practical suggestion. Before you debate whether to hire entry-level employees, first make a list of the training functions that grunt work used to serve in your organization. That list will reveal exactly what needs to go into the learning system your organization must redesign for the AI era.

Closing

When AI takes over entry-level tasks, we also need to examine what learning opportunities those tasks used to provide. The companies featured in this piece are rebuilding the paths through which new hires learn — AI-based training, coaching from seniors, and in-person collaboration. This is an experiment worth studying for Korean companies that have relied on jung-go-sinip, or “experienced rookies” — job-hoppers hired into entry-level roles who already come with some work experience.

What was the first thing new hires learned in your organization? If that task has disappeared since AI adoption, I’m curious what’s filling the gap now. I’d love to hear from you either way — whether you’re a new hire navigating this shift yourself, or someone grappling with how to train one. Share your experience in the comments.

---

📨 If you know a colleague or team lead wrestling with entry-level hiring, please share this piece with them.

---

The draft looks accurate and complete. No corrections needed.

References & Further Reading

Primary sources

• Brynjolfsson, E., Chandar, B., & Chen, R., “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, Stanford Digital Economy Lab, 2025. (PDF) ··· This paper documents the employment decline in AI-exposed occupations among 22-to-25-year-olds. It’s the data that started this whole piece, so if you only read one source, make it this one.
• PwC, “Two Futures for Jobs in an AI Era: 2026 Global AI Jobs Barometer”, 2026. (Full report PDF) ··· Using 1 billion job postings, it shows the “seniorization” of entry-level listings. Worth a look especially if you’re the one drafting job postings.
• Ramp Economics Lab & Revelio Labs, “A New Look at AI’s Impact on Jobs”, 2026. (Revelio Labs commentary) ··· A counterintuitive finding: companies that invest more heavily in AI actually hire more entry-level workers. Read it alongside the research team’s own caveat that this isn’t causation, or you’ll come away with the wrong idea.
• Dell’Acqua, F. et al., “Navigating the Jagged Technological Frontier”, Harvard Business School Working Paper No. 24-013, 2023. ··· The original paper behind the experiment with 758 BCG consultants. This is the source of the “jagged frontier” concept — the uneven boundary between what AI does well and what it doesn’t.
• NACE, “Employers Expect to Hire 5.6% More New College Graduates This Year”, 2026. ··· A survey showing US new-grad hiring outlooks flipping within half a year. Keep in mind the sample is just 185 employers.
• Financial Times, “AI isn’t destroying entry-level jobs. It’s changing them”, 2026. ··· The article that inspired this piece. It goes into detail on the Pininfarina, DHL, and Cognizant cases. (FT subscribers only)

Background

• Pearson & AWS, “AI Readiness: Building the Bridge from Higher Education to Work”, 2026. ··· Shows the perception gap between universities (78%) and employers (53%) across six countries. If you work in education, the framework section is useful.
• Incruit, “More ‘mid-level rookies’ on the rise… 61% of workers say they’d apply as entry-level, career be damned”, 2025. ··· The source of the Korea-section figures in today’s issue.

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

Footnotes

1. Tacit knowledge: knowledge that isn’t written down in documents or manuals and can only be passed on through experience. Think of it as the kind of know-how you learn on the job — like knowing “you have to phrase it this way with this particular client.” ↩