Everything About AI
NextFin News - The US information sector shed roughly 23,000 jobs in August, the largest single-month decline in a three-year slide that has erased about 227,000 positions since the sector peaked in January 2023. The losses are not random. They are concentrated in the roles that once served as the training ground for an entire generation of technology workers - junior developers, data analysts, and entry-level content and infrastructure jobs - and the pattern points to a structural change in how firms build talent, not a cyclical downturn that will heal on its own.
The August 2026 employment report, released September 4 by the Labor Department, delivered a number that landed quietly but carried a loud message: information-sector employment fell by 23,000 in a single month, nearly triple the average monthly loss of about 8,000 recorded over the preceding 12 months. Within that decline, the subsectors most exposed to generative AI took the hardest hits. Computing infrastructure providers, data processing, web hosting and related services lost 8,000 jobs; publishing industries lost 7,000; broadcasting and content providers lost 5,000. The report also showed total nonfarm payrolls rising by 162,000 and the unemployment rate holding at 4.1 percent - a broad market that remains resilient even as the technology-adjacent core hollows out.
This is not a one-month anomaly inside a healthy trend. The information sector - which spans technology, media, publishing, telecommunications, and data services - peaked at 3.054 million seasonally adjusted jobs in January 2023, the month generative AI entered mainstream awareness with the public release of ChatGPT. By April 2026, the latest data point available before the August release, the sector stood at 2.827 million, a net loss of 227,000 jobs over 39 months. The sector has never regained its pre-pandemic employment level and sits roughly 1.8 percent below its February 2020 baseline.
The same seniority-skewed pattern shows up in vacancy data, not just payrolls. Research from Northeastern University tracking 575,397 individual job vacancies for software developers, aggregated into 54,456 job-title-by-month-by-location cells, found that by November 2023 - one year after ChatGPT's release - the relative proportion of junior-level vacancies had fallen 4.7 percentage points below its pre-release trend line. Senior-level postings, by contrast, ran 1.6 percentage points above trend, and openings with no experience level specified ran 3.1 percentage points above trend. A separate study of entry-level software work estimated a 14 to 15 percent relative decline in junior versus senior software developer vacancies after the model's public release.
The mechanism is visible in the hiring data. Stanford Digital Economy Lab researchers, using high-frequency administrative payroll data from ADP covering millions of US workers through June 2026, found that employment of workers aged 22 to 25 in the most AI-exposed occupations now stands 19 percent below where it would be had it kept pace with less-exposed peers. Experienced workers in those same occupations show no comparable gap. The divergence, first documented in August 2025, has widened steadily. Critically, it operates primarily through reduced hiring of young workers rather than increased separations - meaning the damage is done not by mass layoffs but by jobs that are never posted.
The signal is not confined to technology. Indeed Hiring Lab reported that as of May 2026, senior-level job postings were up 14.7 percent year over year while entry-level postings were down 7.5 percent. Compared with January 2025, senior postings were up 13.5 percent, mid-level postings down 6.7 percent, and entry-level postings down 6.3 percent. Yet even after that shift, only about 14 percent of postings nationwide are for senior positions, while nearly half - 46 percent - remain entry-level. The ladder is not disappearing; its bottom rungs are being sanded down.
The Mechanism: AI Substitutes for the Tasks That Used to Train Juniors
The first question is not whether AI is taking jobs. It is which tasks AI takes, and what those tasks used to be for. For decades, the economics of professional services - software, consulting, finance, law, media - rested on a simple bargain: juniors do the routine, repetitive work; seniors review it; and in the process of being reviewed, juniors become seniors. Code debugging, data cleaning, first-draft reporting, basic financial models, document review. These were not merely low-value tasks to be offloaded. They were the curriculum.
Generative AI attacks exactly that curriculum. A model that can draft a function, summarize a document, build a pivot table, or write a first pass of marketing copy does not just reduce the hours a junior spends on a task. It removes the repetition through which judgment is formed. When one senior augmented by AI can produce output that previously required one senior plus three juniors, the firm saves money immediately - and the three juniors are never hired, never interviewed, never posted.
This is why the vacancy data matters more than the layoff headlines. The Stanford researchers found the divergence operates through reduced hiring rather than increased separations. Harvard researchers studying 62 million workers across 285,000 US firms from 2015 to 2025 reached the same conclusion: the decline in junior positions at AI-adopting firms is driven primarily by slower hiring, not by firing. The phenomenon has a name in the economics literature: seniority-biased technological change. Technology has historically favored the skilled. This generation favors the experienced specifically because the inexperienced no longer have a task portfolio that cannot be automated.
The fairness dimension sharpens the problem. When the bottom rungs narrow, the impact falls unevenly on first-generation professionals, career changers, and workers without elite networks - the people who depend on structured entry points rather than referrals. The underemployment rate for recent college graduates rose to 41.8 percent in the third quarter of 2025, the highest level since 2020, according to the Federal Reserve Bank of New York. Employers project only a 1.6 percent increase in hiring for the class of 2026 compared with the class of 2025, and 45 percent of them characterize the job market for new graduates as merely "fair" - the weakest collective rating since 2021.
Why It Did Not Break the Labor Market - Yet
The honest counterweight is that the broader labor market has not collapsed. The August report showed total nonfarm payrolls rebounding with a gain of 162,000 jobs, and the unemployment rate holding at 4.1 percent. The Stanford authors themselves describe their findings as "canaries in the coal mine" - early, descriptive indicators rather than evidence of economy-wide displacement. Their first documented fact is that there is no evidence of widespread job loss across the economy.
That distinction matters. A cyclical reading would say: the information sector over-hired during the 2021-2022 boom, and this is a normal correction. Some support exists for that view. The sector overshot its pre-pandemic trend, and a portion of the decline is simply the reversal of pandemic-era over-staffing. Interest-rate increases also hit the rate-sensitive technology sector hard, independent of AI.
But three pieces of evidence push against a purely cyclical reading. First, the decline is seniority-skewed, not sector-skewed alone - within the same occupations, young workers fall while experienced workers hold steady or grow. A cyclical downturn would not discriminate by years of experience so cleanly. Second, the divergence has widened steadily over nearly three years rather than stabilizing, and it persists when researchers exclude technology firms and computer occupations entirely. Third, the decline is concentrated in occupations where AI usage primarily substitutes for human tasks; where AI usage primarily complements workers, employment is flat or rising, especially for experienced workers. That task-level granularity is the fingerprint of a technological mechanism, not a demand cycle.
The right call is that both forces are present, but they operate on different time horizons. The cyclical leg - over-hiring reversal, rate sensitivity - explains the level of employment. The structural leg - the seniority skew, the task substitution pattern - explains the composition. Levels can recover. Composition changes do not revert on their own.
The Second-Order Consequence: The Pipeline Paradox
The first-order effect of AI-augmented seniors is lower labor cost and higher output per worker. That is the part firms price in, and it is real. The second-order effect is what almost no hiring plan accounts for: if juniors are not hired, mid-level workers do not materialize three to five years later.
Entry-level roles exist for two purposes: to produce output and to produce experienced workers. AI eliminates the economic justification for the first purpose, which destroys the second. The talent shortage that executives fear is not a current shortage; it is a future one, and it is being manufactured today by perfectly rational cost-cutting decisions. A senior-plus-AI team that replaces one senior plus three juniors looks like a productivity gain on this year's income statement. It looks like a succession gap on the balance sheet of 2031.
The World Economic Forum has framed the risk in similar terms: removing entry-level roles weakens the talent and succession pipeline needed to sustain future growth. Randstad's chief human resources officer, Myriam Beatove, put the trade-off plainly - the short-term cost savings of eliminating junior roles create long-term organizational fragility, and the senior talent shortage will not appear for five to ten years, by which time it will be too late to fix quickly.
The Counter-Thesis, and What Would Falsify It
The strongest case against the structural reading is straightforward: this is a productivity boom in disguise, and history will absorb the shock the way it absorbed every previous automation wave. Optimists point out that AI is also creating new entry-level-adjacent roles - prompt engineering, model evaluation, AI quality assurance, automation workflow management - and that AI frees some juniors to take on higher-level work earlier. LinkedIn's chief economic opportunity officer, Aneesh Raman, framed the disruption in exactly these terms in a 2025 opinion piece, likening it to the decline of manufacturing in the 1980s:
Now it is our office workers who are staring down the same kind of technological and economic disruption. Breaking first is the bottom rung of the career ladder.
Even he, a bearer of the warning, argues businesses are not doing away with entry-level work altogether and that AI has freed some junior employees to take on more advanced work earlier.
The optimists are not wrong that adaptation is possible. They are wrong about the default trajectory. Manufacturing's hollowing-out was followed by decades of regional and worker-level scarring that policy never fully repaired; the analogy is not reassuring. And the new AI-adjacent roles are far fewer in number than the routine roles they replace, and they demand a hybrid skill set - domain knowledge plus AI fluency - that a new graduate typically does not yet have. That is the trap: the jobs that used to teach the domain knowledge are the ones being automated.
The falsifying signal is specific and observable. If, over the next four quarters, the ratio of junior-to-senior vacancies in AI-exposed occupations returns to its 2021-2022 trend while employment of workers aged 22 to 25 in those occupations closes to within 5 percent of their less-exposed peers, the structural thesis is wrong and this was a cyclical over-hiring correction amplified by a one-time productivity shock. If instead the 19 percent gap for young workers widens beyond 25 percent while senior employment in the same occupations continues to grow, the seniority-biased regime is confirmed and the pipeline problem moves from risk to fact.
Conclusion: Who Benefits, Who Is Exposed, and What to Watch
The beneficiaries and the exposed are already visible. In the short term, the winners are firms that can substitute AI-augmented senior labor for junior headcount - software companies, consultancies, media and publishing operations, and the data-center and cloud infrastructure providers whose own employment is paradoxically shrinking even as demand for their capacity grows. The exposed are new graduates, career changers, and the mid-tier institutions - regional consultancies, local newsrooms, second-tier software houses - whose business model was built on billing junior hours.
Split by time horizon, the picture is asymmetric. Short term - the next six to twelve months - the labor market is likely to stay resilient enough that the story remains a sectoral one, not a macro one; the August payroll rebound and a 4.1 percent unemployment rate give the Federal Reserve room to look through the information-sector weakness. Medium term - two to four years - the mismatch becomes visible in hiring data as companies that cut junior cohorts find themselves short of promotable mid-level talent, and wage pressure for experienced workers intensifies even as entry-level underemployment persists. Long term - five years and beyond - the question is whether firms rebuild the apprenticeship function that the market used to provide for free, or whether the first rung of the ladder remains missing and a generation of workers is permanently diverted into lower-skilled work.
Three scenarios frame the path. The base case: the seniority skew persists at current levels, entry-level underemployment stays elevated in the low 40s, and firms gradually reintroduce structured training programs only after mid-level shortages become painful - around 2029-2030. The upside case: AI creates enough new junior-adjacent roles, and enough companies redesign onboarding around AI-augmented juniors, that the junior-to-senior vacancy ratio stabilizes within two years. The downside case: the 19 percent youth employment gap in AI-exposed occupations widens past 25 percent, underemployment for recent graduates climbs toward 50 percent, and the ladder's bottom rungs do not return for a decade.
What to watch, concretely: the quarterly junior-to-senior vacancy ratio in software and data occupations; the employment level of workers aged 22 to 25 in AI-exposed occupations relative to their less-exposed peers; and the share of entry-level postings that require two or more years of experience - a clean proxy for whether the first rung is being removed. The next updates to the ADP-Stanford Canaries dashboard and the Labor Department's data on youth employment will move the needle.
The first rung of the tech career ladder is not being cut by layoffs. It is being removed by jobs that are never posted - and a ladder with a missing bottom rung does not grow back when the cycle turns.
Explore more exclusive insights at nextfin.ai.
