TL;DR: A February 2026 Dallas Federal Reserve analysis finds that AI is doing two opposite things at once — eliminating entry-level positions in heavily automated sectors while raising wages for experienced workers who hold tacit knowledge AI cannot replicate. Entry-level roles in the top AI-exposed industries declined 16% as wages for positions in computer systems design rose 16.7%, against a 7.5% national average. Workers aged 22 to 25 are bearing the largest employment losses. The traditional career ladder — enter junior, build tacit knowledge through proximity to experienced colleagues, move up — is losing its bottom rung.

• Entry-level positions in the top AI-exposed industries fell 16% as of early 2026, per Dallas Fed research.
• Workers aged 22-25 saw the steepest employment declines — roughly 20% in heavily AI-exposed sectors in comparable OECD economies between 2023 and 2025.
• Wages in computer systems design rose 16.7% since fall 2022; the national average over the same period was 7.5%.
• AI automates codifiable knowledge — textbook, procedural tasks — while rewarding tacit knowledge built through years of practice.
• The traditional apprenticeship model is the structural casualty firms have not publicly acknowledged.
• California's AB 1898 and AB 2027 address AI notice and worker-data use; there is no federal framework.

AI Wage Gap Entry-Level Workers: The Dallas Fed Data Companies Are Not Advertising

Here is the question few earnings calls address honestly: if AI is making companies more productive, who captures the gains? A February 2026 analysis from the Federal Reserve Bank of Dallas gives the clearest data yet, and the answer is not the people just starting their careers. The study finds AI is simultaneously substituting for junior workers and augmenting senior ones — not as a temporary adjustment but as a structural bifurcation in how labor markets price experience.

Entry-level positions in the top-decile AI-exposed industries dropped 16% in the period studied. Wages in computer systems design — a sector with deep AI exposure — rose 16.7% since fall 2022, more than double the 7.5% national average. The workers aged 22 to 25 recorded the steepest job losses. Ireland's labor data cited in the broader research context showed a 20% employment decline for younger workers in AI-exposed occupations between 2023 and 2025, against 12% growth for prime-age workers aged 30 to 59. The divergence is not subtle.

This matters beyond the raw numbers. The career ladder that built the middle class in knowledge work — enter at the bottom, absorb tacit knowledge through proximity to experienced colleagues, move up — is breaking at its lowest rung. What replaces it is not yet visible in the data.

What "Codifiable" vs. "Tacit" Knowledge Actually Means for Your Paycheck

The Dallas Fed's framework is worth unpacking because it cuts through the usual noise about which jobs AI will or won't take. The core distinction is between codifiable knowledge — skills that can be written down, documented, turned into a procedure — and tacit knowledge, which is learned through practice, judgment built from failure, contextual reading of situations that resists documentation.

AI automates codifiable knowledge with increasing efficiency. Legal research protocols, standard financial modeling, routine code debugging, basic data analysis — these are procedural. The median experience premium across occupations is 40%, the Dallas Fed notes, but the range is extreme. Fast-food workers and ticket agents carry experience premiums below 10%. Lawyers, insurance underwriters, and credit analysts carry premiums above 100%. AI's displacement pressure maps almost exactly onto that distribution: low-premium jobs face substitution; high-premium jobs see augmentation. The math is not subtle, but it rarely appears in company press releases.

The uncomfortable implication is that workers who never get the chance to build tacit knowledge — because entry-level roles are disappearing before they can learn in them — face permanent disadvantage, not temporary disruption. You cannot develop experiential judgment in a role that no longer exists.

Who Benefits When Entry-Level Roles Disappear?

Firms benefit, in the short term. IBM announced it would triple entry-level hires in 2026, which looks generous until you read the context: the company sees AI as a way to compress the development timeline, not extend the workforce. Anthropic's Boris Cherny stated publicly that the foundational software engineering entry-level title could be "extinct by the end of 2026." The savings on junior salaries, benefits, and training overhead go directly to margins or to compensation for experienced workers who remain indispensable.

The OECD's AI-WIPS conference, opening March 30, 2026, specifically added "the rise of agentic AI" and "the governance of algorithmic management" to its formal agenda — a signal that institutions tracking labor markets see this as a durable structural shift, not a cyclical blip. That conference brings together policy, business, academia, and civil society; the fact that worker-side governance now sits alongside productivity metrics reflects how quickly the conversation has moved since 2024.

For workers, the arithmetic is straightforward: tacit knowledge is the only durable competitive advantage against AI substitution. The problem is that tacit knowledge requires codifiable-knowledge apprenticeship as its prerequisite. Remove the apprenticeship stage and you interrupt the pipeline that produces the experienced workers who currently command premium wages. Firms have not solved this. Most have not tried.

The Surveillance Dimension Nobody Puts in the Press Release

The Dallas Fed study focuses on wages and employment counts, but a parallel story runs through the same AI-exposed workplaces. Research published in March 2026 by Race, Power and Policy found that gig workers in AI-managed environments earn a median of $5.12 per hour after expenses — 70% below living-wage standards. Algorithmic management is not just replacing entry-level jobs; it is also degrading the conditions and pay of the jobs that remain.

A pointed international comparison: 83.3% of German call-center agents working in unionized environments with AI tools reported low disciplinary use of performance data. Among non-unionized U.S. call-center workers facing the same AI systems, that figure was 23.2%. The technology is identical in both countries. The power relationship is not. UNITE HERE in Las Vegas secured employer notification requirements before AI deployment; the Transport Workers Union of America won veto power over autonomous vehicle deployment in Columbus, Ohio. These outcomes are not technology achievements. They are bargaining outcomes, and they are only available to roughly 10% of the U.S. workforce — the share that still belongs to a union.

The Legislative Response: State-by-State Patchwork vs. the Scale of the Problem

The March 20 White House "National AI Legislative Framework" established recommendations for Congress to develop a unified federal approach but imposed no new employer obligations and set no enforcement timeline. The gap between state-level momentum and federal inaction means protection is uneven by design. A worker in California has statutory AI notification rights a worker in Texas does not. Neither has federal protection. The Troutman Privacy AI law tracker compiles the weekly state-level bill movements if you need to follow this closely.

When You Should NOT Assume AI's Wage Effects Are Neutral

Several narratives currently circulate that deserve scrutiny. A Stanford working paper found AI "raises average wages by 21% while substantially reducing wage inequality" — a result that received significant media attention. The mechanism cited is AI simplification enabling workers across skill levels to compete for the same jobs. That finding conflicts with the Dallas Fed's occupation-level wage data, which shows the opposite at the bottom of the wage distribution: a strong negative relationship between AI adoption rates and wage growth for lower-paid roles.

Averages mislead here. If wages for senior engineers rise 30% and wages for junior workers fall 5%, the average looks fine. The distribution does not. The IMF's research on AI adoption and inequality found that AI exposure is positively correlated with income at the top of the distribution and negatively correlated with wage growth at the bottom — meaning the workers who most need productivity gains from AI are the least positioned to capture them. That finding predates 2026 but has not become less relevant.

• ☐ Ask whether wage data is disaggregated by experience quartile before accepting "AI raises wages" claims.
• ☐ Check whether employment count trends accompany any wage analysis — wage growth in a shrinking occupation is a different story from growth in a growing one.
• ☐ Distinguish between AI-augmented sectors (rising wages, stable or growing employment) and AI-substituted sectors (rising wages at senior levels, falling employment at entry level).
• ☐ Identify the power structure: unionized workplaces show systematically different outcomes from non-unionized ones facing identical AI deployments.
• ☐ Ask what happens to workers displaced from entry-level AI-exposed roles — do they move to adjacent sectors, leave the labor force, or cycle into lower-quality AI-managed gig work at $5.12/hour after expenses?

What the OECD's Agenda Signals About Where Policy Is Heading

The OECD AI-WIPS conference agenda for March 30 to April 1, 2026 is a reasonable proxy for where the international policy community thinks the pressure points are. The explicit inclusion of "algorithmic management governance" alongside productivity and skills signals that institutions are moving past the question of whether AI disrupts labor markets and toward the harder question of who bears the cost when it does. Germany's active funding of the programme is worth noting — Germany's co-determination tradition, where workers hold board seats at major companies, creates structural incentives to negotiate AI deployment rather than simply impose it.

That model has limited direct applicability to U.S. labor markets, but it demonstrates that the wage divergence the Dallas Fed documents is not technologically determined. The technology is available in both countries. The outcomes differ because the institutional structures differ. That is, in some sense, hopeful: if the split is institutional, it can be changed through institutional means. The question is whether that happens before the entry-level pipeline is too depleted to recover.

FAQ

What does the Dallas Fed data actually show about AI and entry-level wages?

The February 2026 analysis found entry-level positions in top AI-exposed industries declined 16% while wages in AI-intensive sectors rose 16.7% — more than double the 7.5% national average. The gains concentrate in experienced workers with tacit knowledge AI cannot easily replicate.

Which workers face the biggest AI employment risks right now?

Workers aged 22-25 have experienced the sharpest employment losses in AI-exposed sectors. Roles requiring primarily codifiable, procedural knowledge — legal research, standard coding tasks, routine data analysis — face direct substitution pressure regardless of whether they are labeled entry-level.

Are there states with enforceable worker protections against AI employment decisions?

California, Colorado, Illinois, and New York City have enacted or are advancing AI employment laws. California's AB 1898 requires employer notice when AI assists in employment decisions. Colorado's law effective June 2026 regulates high-risk AI in hiring and promotion. No federal law currently applies to most U.S. workers.

Is the Stanford finding that AI reduces wage inequality credible?

The finding rests on a specific mechanism — AI simplification enabling cross-skill competition — that does not appear in occupation-level wage distribution data from the Dallas Fed. Averages can obscure widening gaps at the tails. Both findings can coexist if AI raises average wages while compressing or eliminating entry-level pathways in specific sectors.

What is algorithmic management and why does it matter for wages?

Algorithmic management uses AI to control scheduling, performance scoring, task assignment, and discipline. Research shows it correlates with wage suppression in non-unionized environments and with gig wages as low as $5.12 per hour after expenses. The governance of these systems is now on the OECD's formal 2026 policy agenda.

Conclusion: Next Steps

The Dallas Fed's February 2026 data makes the shape of AI's labor market impact concrete: wages are rising where experience is irreplaceable and falling — or jobs are disappearing — where codifiable knowledge used to be enough. This is not a projection about 2030. It is a description of what wage and employment data already show in 2026.

The policy response is fragmented. State legislatures are moving; the federal government is producing frameworks without mandates. The OECD is convening international consensus-building at its AI-WIPS conference. None of this operates at the speed of AI deployment. Firms are capturing productivity gains faster than workers are gaining legal protections.

If you work in a profession where your value is predominantly procedural — where your knowledge could, in principle, be written into a manual — the data suggests the time to build tacit, judgment-intensive expertise is now, not after the next round of layoffs. If you set policy, the question the Dallas Fed data raises is direct: who is responsible for the workers who lose the entry-level positions that used to produce the experienced workers organizations still need?