Who's Getting Hit Hardest, and Why Experience Matters More Than You'd Expect
This is the fifth in a series exploring what the research actually says about AI and your career. Previously: AI displacement is real, which roles are most exposed, AI is quietly rewriting your role, and is your company actually restructuring?. This week: who's bearing the brunt of all this, and why experience matters more than you'd expect.
Who's getting hit hardest, and what both early-career and experienced professionals can do about it
Over the past four weeks, we've built a picture of how AI is reshaping work. Displacement is real. Which roles are most exposed depends on what you actually do every day. AI edits your role by taking specific work tasks first, and the edit can either simplify your role (deskilling) or make it more valuable (upskilling). And most companies have adopted AI tools but haven't yet reorganized around them.
This week's question brings it down to people: who's bearing the brunt of all this, and what can both early-career and experienced professionals actually do about it?
The data is now clear: AI hits harder at the bottom of the experience ladder
Two independent research teams, using completely different data and methods, arrived at the same conclusion.
A Stanford team (Brynjolfsson, Chandar & Chen, 2025) analyzed millions of ADP payroll records and found that people aged 22-25 in AI-exposed roles saw employment drop 6-16% since late 2022. Young software developers specifically declined nearly 20% from their late-2022 peak. Meanwhile, people over 30 in those same roles saw employment grow 6-12%.
A Harvard team (Hosseini & Lichtinger, 2025) analyzed 62 million LinkedIn profiles and 200 million job postings across 285,000 companies and found the same pattern from a different angle. At companies that actively adopted GenAI, junior hiring declined sharply while senior hiring continued to rise.
The fact that two separate teams, from two different universities, using two entirely different datasets (payroll records vs. LinkedIn profiles), independently found the same pattern is strong evidence. This isn't one study that might be wrong. It's a convergence.
The Hosseini & Lichtinger paper gives this phenomenon a name: seniority-biased technological change. Prior automation waves favored educated people over less-educated people (skill-biased change). AI is doing something different: it favors experienced people over inexperienced people, regardless of education level. A 35-year-old and a 23-year-old with the same degree face different outcomes.
The numbers around it are striking. Entry-level tech hiring dropped 73% year-over-year (Ravio, "2025-2026 Tech Job Market and Compensation Reports"). Entry-level job postings overall are down 35% since January 2023 (Revelio Labs, cited in CNBC, Aug 2025). Unemployment for recent college graduates hit 5.6% in Q4 2025, with 42.5% of graduates working in roles that don't require a degree (NY Fed). This past September, Fed Chair Powell directly acknowledged the pattern: "You are seeing some effects... A particular focus on young people coming out of college."
Takeaway
AI displacement isn't hitting everyone equally. It's concentrated among early-career people in AI-exposed roles. Experienced professionals in those same roles are not just surviving but growing. The seniority bias is real, measurable, and confirmed by independent research.
What you learned in school vs. what you learned on the job
The mechanism behind the seniority bias is straightforward once you see it.
Early-career people bring mostly codified knowledge: what you learn in school, from textbooks, from training programs. How to write code in Python. How to build a financial model in Excel. How to draft a contract following standard clauses. This knowledge has clear inputs, defined outputs, and can be described as a set of instructions. Which is exactly the kind of work AI handles well.
Experienced people bring something different: tacit knowledge. The judgment calls that come from years of doing the work. Knowing which stakeholder will push back on a proposal and why. Understanding that the code compiles fine but the architecture will create scaling problems next quarter. Sensing that a customer's real concern isn't what they said in the meeting. None of this is written down anywhere. None of it fits neatly into a prompt.
AI can do the codifiable parts of a role. It struggles with the tacit parts. So companies are making a rational calculation: hire fewer people at the entry level where the work is mostly codifiable, and keep (or hire more) experienced people whose tacit expertise AI can't replicate.
The tacit layer is thinner than you think
As we covered in the AI exposure map, agentic AI is pushing this boundary further. Agentic systems can chain steps, maintain context across a workflow, and self-correct, which means they can now handle coordination that used to feel like it required human judgment: knowing which step to do first, what to check before moving to the next phase, when to loop back. Much of that coordination turns out to have been implicitly codified: it followed patterns that nobody had written down but that an agent framework can learn to execute. What remains genuinely tacit is knowing when the standard workflow is wrong for this specific situation, reading a client's unstated concern, sensing that a technically correct output will fail for organizational or political reasons. Agentic AI is eating into what people thought was tacit but was actually codifiable coordination. The truly tacit layer remains human, but it's a thinner layer than most people assume.
The Harvard study found something important about how this plays out inside companies. It's not that companies are laying off junior people. They're simply not hiring them. Open positions aren't being posted. Roles aren't being backfilled when people leave. And within remaining junior roles, the specific work tasks most exposed to GenAI are being explicitly removed from job descriptions. The signal is quiet: no layoff announcements, no WARN notices. Just a gradually thinning pipeline.
One more finding from the Harvard study that surprised researchers: the impact follows a U-shaped pattern by education tier. Picture the education spectrum from left to right: elite universities on one end, mid-tier schools in the middle, non-degree paths on the other end. Now plot how much each group's employment declined. The curve dips at both ends and peaks in the middle, forming a U. In other words, mid-tier graduates bear the most displacement.
(Ivy, Stanford, MIT)
international
& regional
standard
selective
Graduates from elite institutions (top-20 type programs) are less affected, partly because their degrees carry signal value (hiring managers infer qualities beyond what's taught in class, like selection through a highly competitive process) and partly because their alumni networks provide referral channels that bypass the AI-screened job application pipeline entirely. Non-degree holders are also less affected, largely because they tend to work in physical or interpersonal roles with low AI exposure. It's the graduates from solid state schools, with standard credentials and no standout network advantage, who are most concentrated in the codified-knowledge roles that AI handles best.
Takeaway
Experience protects you because AI can replicate what you learned in school but not what you learned from doing the work for years. But this protection is conditional. It holds as long as your tacit expertise stays ahead of what AI can do. If you stop growing, the boundary moves past you.
The deeper problem nobody's talking about: the apprenticeship pathway is breaking
Here's the part that goes beyond "entry-level people are struggling" into something more structural.
Entry-level work was never just cheap labor for companies. It served a dual purpose. Yes, it provided necessary output (the reports got compiled, the code got written, the data got entered). But it also provided a training mechanism. By doing that structured, codifiable work alongside more experienced colleagues, early-career people absorbed tacit knowledge through observation and practice. They learned which numbers could be trusted and which couldn't. They saw how production systems failed. They discovered how decisions actually got made, as opposed to how the org chart said they should.
Enrique Ide, an economist at IESE Business School, formalized this in a recent paper (April 2026). His model shows that automating entry-level work can increase output today while reducing growth and capability over time, even without reducing the total number of jobs. The mechanism: if early-career people don't do the work, they don't absorb the tacit knowledge. If they don't absorb it, they can't become the experts of the next generation. Organizations gain short-term efficiency while eroding the pipeline that produces their future senior talent.
Think of it this way. A junior analyst who spends two years building financial models learns which assumptions break under stress, not from a textbook but from getting it wrong and being corrected by someone who got it wrong ten years earlier. That correction, that specific "here's why this assumption will fail when interest rates spike," is tacit knowledge being transferred. AI can build the model faster. But AI doesn't teach the junior analyst what to distrust about the model's output.
This creates a genuine catch-22 for new graduates. The standard advice is: "develop judgment, interpersonal skills, domain expertise." But where do you develop those? At work. Doing the entry-level work that AI just took. The pathway to acquiring tacit knowledge required doing codifiable work alongside experts. AI removed the codifiable work, and with it, the vehicle for learning.
LinkedIn's chief economic opportunity officer, Aneesh Raman, likened it to the decline of manufacturing in the 1980s: "Breaking first is the bottom rung of the career ladder."
Takeaway
The entry-level problem isn't just about today's graduates finding jobs. It's about whether organizations are preserving the pathway through which people develop the expertise that AI can't replicate. If that pathway breaks, everyone loses, including experienced professionals who depend on the next generation of talent.
What early-career people are actually doing about it
The data shows three main responses:
1. Building their own path.
Nearly 38% of 2025/2026 graduates are considering starting their own business. Another 32.5% are looking at gig work, 28% at freelancing, and 11% at skilled trades (ZipRecruiter survey). Over half of Gen Z professionals already freelanced in 2023, the highest of any generation (Upwork). New business applications remain at elevated post-pandemic levels.
The irony is notable: the same AI tools that eliminated entry-level work have collapsed the cost of starting a venture to near zero. You can build a product, run marketing, handle customer service, and manage operations with a few hundred dollars and the right AI stack. Whether this path builds the long-term tacit expertise that employers value is an open question, but it's clearly where a growing share of new graduates are heading.
2. Targeting companies that redesigned entry-level roles rather than eliminating them.
Not every company responded to AI by cutting junior hiring. IBM is tripling its Gen Z hiring and rewriting every entry-level job description for the AI era (Fortune, Feb 2026). Software engineers at IBM now spend less time on routine coding and more on customer interaction. HR staff focus on chatbot oversight rather than answering every employee question. Their CHRO explicitly warned that cutting entry-level talent backfires in the long term.
Cognizant hired 25,000 fresh graduates in 2025 and expects to exceed that in 2026, despite extensive AI use across the company (WEF, March 2026). Amazon is bringing on 11,000 software engineering interns in 2026. AWS CEO Matt Garman said: "We are hiring just as many software developers as we ever had."
These companies share a common approach: they redesigned the entry-level role rather than eliminating it. The new version emphasizes AI output review, customer interaction, exception handling, and workflow monitoring. These are durable skills that build tacit knowledge faster than the old "compile the data and draft the memo" model ever did.
3. Treating AI fluency as the entry ticket, not the differentiator.
84% of developers now use AI tools (Stack Overflow 2025). AI literacy is the fastest-growing skill on LinkedIn. People with AI skills earn a 56% wage premium, doubled from 25% a year prior (PwC 2025). But AI fluency alone isn't enough. As MIT Technology Review argued, "The competition most young workers will experience is not human versus machine but colleague versus AI-augmented colleague." The people who stand out combine AI fluency with a domain (the mechanical engineer who knows manufacturing and AI, the finance grad who understands compliance and AI workflows). AI fluency is table stakes. The domain depth is the differentiator.
Takeaway
Early-career people who are adapting successfully aren't waiting for traditional entry-level roles to come back. They're building their own experience through ventures and freelancing, targeting redesigned roles at companies like IBM and Cognizant, and combining AI fluency with domain specialization.
What mid-career professionals should understand about their protection
If you're 5-15 years into your career, the data currently works in your favor. You're on the protected side of the seniority bias. Your tacit knowledge, your relationships, your judgment under ambiguity: these are exactly what AI can't replicate.
But the protection comes with conditions.
Condition 1: You have to keep building tacit expertise, not just rely on what you've already accumulated.
The boundary between "what AI can do" and "what requires human judgment" keeps moving. Tasks that felt like they required judgment two years ago (research synthesis, first-draft analysis, vendor comparison frameworks) are now handled by AI tools. What's tacit today may become codifiable tomorrow as AI capabilities expand. Ide's research makes this point formally: your tacit knowledge depreciates if you stop doing the judgment-intensive work that builds new expertise. If you delegate all the hard problems to AI and spend your time reviewing AI output, you eventually stop growing. Your current expertise becomes stale.
Condition 2: The "AI-augmented experienced professional" is the new competitive baseline.
Research consistently shows that AI makes experienced people more productive. Professionals with strong AI knowledge are 2.8 times as likely to see organizational benefits (Thomson Reuters, 2025). The 56% wage premium for AI-skilled professionals applies at every experience level (PwC, 2025). What this means in practice: you're no longer competing against peers who work without AI. You're competing against peers who work with AI. The experienced professional who learns to direct AI tools effectively produces more, faster, and at higher quality. The one who doesn't is at a growing disadvantage, regardless of experience level.
Condition 3: Your expertise becomes more valuable when you encode it into AI workflows, not less.
This is counterintuitive. You might think: if I feed my tacit knowledge into AI systems, won't I make myself replaceable? The research suggests the opposite. McKinsey's work on how organizations are evolving around AI (September 2025) found that specialists who encode their domain expertise into AI workflows gain influence. The accountant who can specify what an AI agent should check for in an audit is more valuable than the one who manually does the checking. The compliance officer who can define regulatory constraints for an automated workflow becomes essential. Your expertise becomes the input that makes AI systems useful. Holding it back doesn't protect you. It just means someone else encodes theirs first.
The reason this works (and doesn't make you replaceable) goes back to Ide's distinction: AI can capture the product of your expertise (what you know right now) but not the process by which you acquired it. The world keeps changing. New situations arise that the encoded rules don't cover. The person who went through the learning process can recognize the gap and update the knowledge. The AI system can't. Your ongoing judgment is what keeps the encoded expertise relevant.
Takeaway
Your experience is currently your strongest asset. But it protects you only if you keep building new expertise, learn to work with AI effectively, and position yourself around the durable parts of your work. The goal isn't to hide your knowledge from AI. It's to keep doing the hard work that builds new knowledge while using AI to clear the routine.
Where to start
Both early-career and experienced professionals share the same first step: understanding which parts of their work (or the roles they're targeting) are exposed to AI and which parts are durable.
For experienced professionals, that's what Alignment Resilience was built for. It maps your specific work against 19,265 human-rated tasks, your actual resume, and live market data. You get a Resilience Score, a task-by-task breakdown, your durable assets with market demand trends, concrete career paths forward, and a 30-day action plan for each path. Whether you use the clarity to reposition internally or to lead your next job search with your AI-resilient strengths, the starting point is the same.
For early-career professionals evaluating which roles to target, Alignment Check shows the task composition of any role, scored by AI exposure.
Research referenced:
- Brynjolfsson, Chandar & Chen, "Canaries in the Coal Mine," Stanford Digital Economy Lab, November 2025
- Hosseini & Lichtinger, "Generative AI as Seniority-Biased Technological Change," Harvard University / SSRN, August 2025
- Ide, "Automation, AI, and the Intergenerational Transmission of Knowledge," SSRN, April 2026
- Noy & Zhang, "Experimental Evidence on the Productivity Effects of Generative AI," Science, 2023
- Brynjolfsson, Li & Raymond, "Generative AI at Work," NBER, 2023
- MIT Technology Review, "It's time to address the looming crisis in entry-level work," May 26, 2026
- Ravio, "2025-2026 Tech Job Market and Compensation Reports"
- Revelio Labs, entry-level job posting data, cited in CNBC, August 2025