The GenAI Paradox, Part 3: The Invisible Workforce Crisis
Junior roles frozen, middle managers becoming therapists, and a generation of CS graduates competing against tools that are faster and cheaper. The workforce impact no one in the C-suite wants to quantify.
This is Part 3 of a five-part series. Read Part 1: The Great Divide | Part 2: Boardroom Hope vs Operational Reality | Part 4: Shadow AI and the Singapore Model | Part 5: Trust, Governance, and What Comes Next
The impact of AI on the workforce is defined by a widening gulf between the “official” narrative of empowerment and the lived experience of anxiety, displacement, and hidden labour.
There is a deeper logic underneath the disruption, and it explains why the pain lands where it does. For two centuries, economic value flowed to whatever could be made explicit, codified, and scaled — and the human was valued only as a scarce input to that machine. Generative AI drives that logic to its limit: it commoditises the last scarce human input to the explicit, the part of the job that was already half-written-down. What it cannot touch is the irreducibly human remainder — earned judgment, accountability, the embodied feel for a problem. So the value of work is quietly inverting away from the codifiable and toward the human-as-human. The workforce crisis below is what that inversion looks like before anyone has named it: the codifiable tier is collapsing, and the institutions that used to build the human remainder are being dismantled at the exact moment they matter most.
The Junior Freeze and the Skills Gap
One of the most alarming trends in 2025 is the collapse of the entry-level job market in technology and knowledge sectors.
The “Codified Knowledge” Trap. AI tools have become proficient at tasks typically assigned to junior employees — coding boilerplate, drafting emails, summarising data. Consequently, companies have frozen hiring for junior roles, preferring to hire seniors who can orchestrate AI agents.
What the trap actually automates is worth naming precisely, because it explains both the speed of the freeze and its hidden cost. The junior tier was never about the output — it was the place where substrate got built: where the boilerplate, the draft, the summary were the means by which a novice slowly earned the judgment that makes a senior. AI is extraordinarily good at exactly the part that was already half-written-down — the codifiable, near-explicit work — which is precisely the work the junior used to climb on. When that work is borrowed from a model rather than learned by a person, the output is real but the substrate underneath it never forms. The novice produces senior-looking work while granting the fluent output an unearned trust — what our research calls Epistemic Credit — and no judgment is laid down to catch it the day it is fluently wrong.
The Data. A Stanford Digital Economy Lab study (Brynjolfsson, Chandar & Chen, August 2025) reveals a 13% relative decline in employment for workers aged 22–25 in AI-exposed occupations, compared to workers in low-exposure occupations. For software developers specifically in that age bracket, the decline was closer to 20%. This creates a “missing rung” in the career ladder — and raises long-term questions about how the next generation of experts will be trained if they cannot enter the workforce.
This is not a quirk of the 2025 tech market; it is an old structural law arriving at unprecedented speed. We call it the Squeezed Middle: wherever mastery is built by climbing a supervised middle tier, removing that tier looks efficient and breaks expert renewal — the failure is simply deferred. It has been seen across crafts, aviation, and medicine; AI is its fastest and highest-reaching instance, because the cheap codification that lets a firm skip the junior also lets it skip every other place a person used to learn by doing. The bill does not come due this quarter. It comes when the seniors who were trained the old way retire and there is no one underneath who built the same judgment.
The Sentiment. Practitioner communities are filled with despair among recent CS graduates who feel they are competing against tools that are faster and cheaper, leading to a pervasive sense of being “scammed” by the educational system that trained them for roles that are contracting.
This is the pipeline problem we identified in our analysis of AI and software engineering: if AI substitution compresses the junior tier, the senior engineer shortage does not show up for five to seven years — long after the executives who made the hiring decisions have moved on.
Middle Management: The Ecosystem Managers
Middle managers are absorbing the shock of this transition, their roles shifting from task oversight to “ecosystem management.”
The “Therapist” Role. Managers report spending increasing time managing the emotions of their teams — fear of layoffs, burnout, cynicism — rather than the work itself. They are the buffer zone between executive AI mandates and workforce reality.
Managing Invisible Work. AI-driven dashboards often present a sanitised version of reality (“everything is fine”), failing to capture the “shadow work” required to fix AI errors or the burnout of employees juggling unrealistic targets. Managers must navigate “invisible negotiations” to keep the system running.
This is a precise instance of a failure our research warns against. The codified store of an organisation — its dashboards, its data, its automated reporting — is what we call the Institutional Vault: it is excellent at holding and executing what humans have already decided, but it has no judgment of its own. The moment leadership lets the green dashboard decide that everything is fine — closing the loop on a call no human actually made — the Vault has been allowed to judge, and the middle manager who can see the real shadow work is the only one left who can override it. That override is not noise in the system; it is the system working. Strip the manager out and the sanitised version becomes the truth of record.
Shadow Automation. Managers also face the challenge of employees quietly automating their work but hiding the productivity gains to avoid increased quotas. This creates a “cat-and-mouse” game that erodes trust and obscures true capacity.
The middle management crisis is one of the least discussed and most consequential dynamics of enterprise AI adoption. These are the people responsible for translating executive AI strategy into operational reality — and they are doing it without support, recognition, or adequate tooling. The capacity they are exercising is precisely the one the inversion makes scarce: the Translator who can hold the codified mandate and the human reality in the same head and reconcile them. It is irreducibly human, structurally undervalued, and being burned out instead of built. That is the moat being eroded in plain sight, mistaken for an HR cost.
Practitioner Sentiment: AI Fatigue
Analysis of practitioner communities reveals deep-seated “AI fatigue” across the technology workforce.
“Forced Integration.” Developers and knowledge workers describe “forced usage” of AI tools that are perceived as surveillance mechanisms or “feature factories” rather than genuine productivity aids.
“AI Slop.” There is growing resentment toward the “echo chamber” effect of AI-generated content, with practitioners describing models degrading into “digital echo chambers of nonsense.”
Skepticism of ROI. Discussions in IT management communities consistently question the ROI of GenAI, describing it as “vaporware” or “hype” that executives have bought into without understanding the technical limitations. This bottom-up scepticism contrasts sharply with the top-down optimism documented in the C-suite sentiment data — and the gap between the two is itself a risk indicator.
This is a pattern the Sovereign Command research calls the “Eloquence Trap” at the organisational level: polished executive dashboards and vendor demos suppress the verification instinct, while the people closest to the operational reality are systematically excluded from the investment decisions.
The Crisis Read Correctly
Put the three together — the frozen junior, the burned-out manager, the fatigued practitioner — and a single picture resolves. Each is a place where the codifiable part of work has been handed to a machine and the human part has been treated as a cost to be cut. But the human part is no longer the overhead; under the inversion, it is the entire asset. The junior tier built judgment. The manager held the line between the dashboard and the truth. The practitioner did the verification the model cannot do for itself. These are not the residue of work that AI is making obsolete — they are the part of work that is appreciating, being dismantled in the belief that it is depreciating.
The mistake is to read the floor going up — AI lifting average output, flattening raw talent — as the human becoming less necessary. It is the opposite. Lifting the floor strips away the false moat of codifiable ability and exposes the real one: the earned, accountable, irreducibly human judgment that the firing decisions are quietly destroying. And that judgment is the one thing that cannot be bought back quickly. It is forged — slowly, by people doing consequential work under supervision — and the same cheap codification that lets a firm skip the junior tier also lets it skip the institution that used to build the substrate. The freeze is not just a hiring decision; it is a decision to stop manufacturing the very capability the next decade will be starved of.
The C-suite does not want to quantify this because the cost is deferred and the saving is immediate. But the bill is real, and it is the moat. The organisations that survive the transition will be the ones that treat the human remainder as infrastructure to be built rather than headcount to be trimmed — that rebuild the missing rung on purpose, because nothing else will.
About the Author
Ethan Seow is a Centre for AI Leadership Co-Founder and Cybersecurity Expert. He is ISACA Singapore’s 2023 Infosec Leader, ISC2 2023 APAC Rising Star Professional in Cybersecurity, TEDx and Black Hat Asia speaker, educator, culture hacker and entrepreneur with over 13 years in entrepreneurship, training and education.
This is Part 3 of a five-part series. Continue to Part 4: Shadow AI and the Singapore Model.