The 2023 forecasts were not wrong — they were incomplete. OpenAI estimated 80 percent of workers had at least 10 percent of their tasks exposed to large language models. Goldman Sachs projected 300 million jobs at risk. McKinsey saw $4.4 trillion in potential value. All three were directionally correct.
What none of them anticipated was that the impact would not be uniform. The workforce is splitting — senior professionals becoming scarcer and more expensive, junior workers facing a structurally blocked entry pipeline, and the middle hollowing out. The gap between “AI could do this” and “AI is doing this” turned out to be the most consequential variable of the past three years. For executive search firms like SunTzu Recruitment, that gap has become the central puzzle: how do you recruit for a labor market that is reorganizing faster than job descriptions can be rewritten?
The Gap Between Theory and Reality
In March 2026, Anthropic released Labor Market Impacts of AI: A New Measure and Early Evidence, drawing a distinction that earlier studies had collapsed. Theoretical AI exposure — the share of tasks a model could technically perform — stood above 80 percent for legal, arts and media, and education. Actual adoption, measured by real API usage, was concentrated in just four domains: computer science, business and finance, management, and office administration.
The gap between the two numbers is not a lag. It is organizational inertia, regulatory friction, and capital allocation preference rolled into one. The BCG-Harvard experiment on the “jagged technological frontier” had already foreshadowed this in 2023: consultants using GPT-4 completed 12.2 percent more tasks and worked 25.1 percent faster — but on complex work beyond the model’s boundary, their performance actually declined. The core skill was not using AI, but knowing when AI would fail.
A recruitment specialist at SunTzu Recruit who tracks AI adoption across China’s technology sector put the pattern bluntly: companies first treated AI as a personal productivity tool, then as a team-level automation layer, and finally as an agent-based operating model. Each stage changed what kind of talent they needed — and the recruitment industry is still catching up to the third stage. From SunTzu Recruitment’s own client engagements, the shift from stage two to stage three happened within roughly nine months during late 2025 — faster than most HR teams could adapt their hiring criteria.
The Learning Curve Trap
Harvard Business School’s Fuller and colleagues, analyzing millions of U.S. online job postings, introduced a variable that changes how we understand AI’s impact: learning curve steepness.
Where the learning curve is steep — senior people dramatically outperform juniors, and compensation reflects that gap — AI automates entry-level tasks first. Junior roles are built from explicit, programmable routines that AI can directly handle. This cuts off the “learning by doing” pathway that historically let new hires accumulate the tacit knowledge required for senior positions. The bottleneck forms at the point of entry, not at the top.
Where the learning curve is shallow — experience carries less premium — AI tends to empower entry-level workers rather than replace them. Novices using AI tools can produce output close to that of seasoned professionals, expanding access rather than contracting it.
The result is what SunTzu Recruit’s senior consultant describes as a gourd-shaped workforce: a thin layer of highly scarce senior talent at the top, a hollow middle, and a blocked pipeline at the bottom. Companies can find senior hires — at steep and rising premiums — but the junior pipeline they depend on for succession planning is drying up. The data confirms a shift that HR leaders had been feeling but could not yet name.
By 2026, the age anxiety threshold in China’s technology sector had moved from 35 to 25. Entry-level candidates born after 2000 are competing not just with each other, but with AI automation of the very tasks that once served as their on-ramp. A SunTzu Recruitment consultant specializing in the AI sector noted that the most common question from client companies has shifted from “Where do we find senior AI talent?” to “How do we rebuild our junior pipeline when the traditional entry roles no longer exist?”
Capital Votes, People Pay the Price
A June 2026 LatePost analysis of major U.S. and Chinese technology companies revealed an unambiguous trend: capital expenditure on computing infrastructure has overtaken R&D spending at most leading firms. The top 20 companies by market capitalization in both countries have tilted decisively toward AI hardware and infrastructure providers over the past five years. The market is voting with real money, and the direction is clear.
Inside these companies, a scissors effect has emerged: revenue per employee rises while headcount stagnates or declines. Growth that once required hiring waves now comes from adding compute. Token consumption has become a new operating cost line — money that previously paid salaries for report writers, copywriters, and data analysts now flows to chipmakers and cloud providers.
A Caixin feature in June 2026 captured the human dimension of this shift. A data analyst built an automation system for her company’s token cost tracking. When the system was completed, her position was eliminated — she moved to traditional manufacturing at a 30 percent pay cut. A short drama operations manager saw her entire Beijing team eliminated when production shifted to AI-generated content; her role was not poorly performed, the business premise it depended on was simply erased. A visual designer returning from maternity leave found that job postings now demanded one person to handle AI image generation, retouching, video editing, and social media operations — the workload of an entire department compressed into a single headcount.
Being inside the AI industry did not grant immunity. It merely meant being first in line.
Searches conducted by SunTzu Recruitment over the past eighteen months reveal a pattern: companies that have successfully restructured their talent strategy around this shift are not hiring to replace the roles eliminated by automation. They are hiring for roles that did not exist in 2023 — AI governance specialists, human-AI workflow designers, autonomous system oversight managers. The job titles are still settling, but the direction is unmistakable.
What This Means for Talent Strategy
That said, the picture is not purely deterministic. Regulatory frameworks in China and the EU are beginning to address the displacement of entry-level roles. Corporate social responsibility pressure is pushing some large employers to maintain apprenticeship-style programs even when automation could replace them. Organizational choice still matters more than technological inevitability.
For SunTzu Recruitment, the practical shift is unmistakable. The traditional candidate assessment framework — years of experience, educational pedigree, technical certifications — is losing predictive power. What matters increasingly is a candidate’s demonstrated ability to work across the human-AI boundary: knowing which tasks to hand to autonomous systems and which to keep under human judgment.
SunTzu has begun incorporating this capability assessment into its executive search methodology. In practice, this means evaluating not just what a candidate has done, but how they have navigated the transition from human-only workflows to hybrid human-AI operations — a distinction that does not show up on any standard résumé.
A partner at SunTzu Recruitment observed that companies still writing job descriptions the same way they did in 2023 are competing with one hand tied. The most effective searches in 2026 start not with a role title but with a task decomposition: which of these responsibilities should an AI agent handle, and which require human judgment? Answering that question well is the difference between a hire that scales and one that is obsolete within eighteen months.
The data from the past three years does not point to a single directional outcome. It points to a set of forces pulling in opposing directions, with organizational choice — and the talent strategy that choice produces — as the deciding variable. For recruiters operating in this environment, the only viable approach is to understand which side of the split each role falls on, and to build strategies that acknowledge the divide rather than pretending it does not exist.
Sources: OpenAI (2023), Goldman Sachs (2023), McKinsey Global Institute (2023), BCG & Harvard Business School (2023), Anthropic Economic Index (2025, 2026), Harvard Business School — Fuller et al. (2025), Stack Overflow Developer Survey (2024), GitHub Octoverse (2024), Upwork (2024), Freelancer.com Quarterly Reports, LatePost (June 2026), Caixin Weekly (June 2026), Tencent Research Institute (July 2026)
Comments are closed