Federal Reserve AI and Coder Employment: Comprehensive Evidence Analysis of Labor Market Disruption

AI Employment Impact Overview and Federal Reserve Findings

A groundbreaking Federal Reserve study has provided the first rigorous quantitative evidence of artificial intelligence’s impact on employment, revealing that coder employment growth has been approximately 3 percentage points lower per year since the introduction of ChatGPT in November 2022.

The research, conducted by Leland D. Crane and Paul E. Soto from the Board of Governors of the Federal Reserve System and released in March 2026, represents the first systematic attempt to measure whether large language models have had a discernible aggregate labor market impact. Using sophisticated econometric methods that link O*NET occupational data to Current Population Survey employment figures, the study tracks monthly employment patterns for coding-intensive occupations.

The methodology centers on an event study design that uses ChatGPT’s November 2022 release as a natural breakpoint for analysis. This approach allows researchers to isolate the effects of AI adoption from other economic factors that might influence employment trends.

“While coder employment has continued to grow post-ChatGPT, it has done so much more slowly than the pre-2022 trajectory would suggest,” the researchers note, emphasizing that this represents a significant deceleration rather than absolute decline.

The study’s significance extends beyond immediate employment figures. It provides policymakers with concrete evidence that AI tools are having measurable economic impacts, moving the discussion from theoretical possibilities to empirical realities. For technology professionals and business leaders, it offers critical insights into how AI adoption is reshaping the most AI-exposed occupation category in the U.S. economy.

Why Coders Are the Canary in the AI Coal Mine

The Federal Reserve researchers chose to focus on programmers for a compelling reason: computer and mathematical occupations account for more than one-third of all Claude AI queries despite comprising only 3.4% of the total workforce.

This disproportionate usage pattern is confirmed across multiple data sources. The Anthropic Economic Index (AEI) data from February 2025 shows coding as the predominant use case for generative AI, with Handa et al. (2025) confirming these queries are “essentially all computer programming-related.” Independent survey data from Bick et al. (2024) and Bonney et al. (2024) consistently identify the highest AI adoption rates among computer and mathematical occupations.

The researchers’ thesis is straightforward: “If generative AI is to substantially affect the job market, the effects should be apparent here first.” This makes programmers an ideal test case for measuring AI’s broader economic impact.

The data supports this hypothesis. Both major AI exposure metrics—the widely cited Eloundou et al. “GPTs are GPTs” study and the Anthropic Economic Index—agree that coders are among the most exposed occupational groups:

• 99.5% of coding employment falls in the top GPTs exposure quintile
• 98.2% falls in the top AEI exposure quintile
• No other major occupational category shows such consistent high-exposure ratings

This concentration of AI usage among programmers creates a natural experiment. If AI tools like ChatGPT, Claude, and GitHub Copilot are genuinely substituting for human labor rather than merely augmenting productivity, the employment effects should be most visible and measurable among coders first.

The choice of coders as a bellwether also addresses a key methodological challenge in AI impact research: measurement timing. Unlike previous automation waves that unfolded over decades, AI adoption has been remarkably rapid. By focusing on the occupation with the highest AI adoption rates, researchers can detect employment effects that might not yet be visible in less-exposed occupational categories.

Defining Affected Occupations: Who Counts as a Coder

The Federal Reserve study provides precise operational definitions for “coding-intensive occupations,” addressing a common challenge in AI impact research. Coding-intensive occupations comprise approximately 3.7% of total U.S. employment, representing about 5.7 million workers as of November 2022.

The researchers used O*NET’s programming skill importance ratings, setting a threshold at ≥2.76 to exclude management-oriented and general engineering roles while capturing occupations where coding is central to job performance.

Top Coding-Intensive Occupations by Employment Share

Software developers represent the largest single group, comprising over 1% of total U.S. employment. While computer programmers have the highest coding importance rating (4.75), they represent a smaller share of the workforce (0.29%). This distinction is crucial for understanding the study’s aggregate findings.

The researchers acknowledge that O*NET’s task descriptions may understate coding intensity for software developers. Self-reported job duties from American Community Survey data confirm that software developers are heavily coding-intensive, even when their formal job titles might suggest broader responsibilities.

The 2.76 threshold was carefully chosen to exclude roles where programming is peripheral. This means the study excludes:

• Engineering managers who occasionally review code
• Product managers with technical backgrounds
• Systems analysts focused primarily on business requirements
• Technical writers who might work with code documentation

The Employment Deceleration: Core Statistics and Trends

The Federal Reserve study reveals a dramatic shift in employment growth patterns for coding occupations. Pre-ChatGPT, coder employment grew at approximately 4.9% annually, significantly outpacing total private employment growth of about 1.3%.

The post-ChatGPT analysis shows striking differences across industry categories:

Employment Growth Changes by Industry Scope (Table 5 – No Controls)

The most striking finding is that in coder-intensive industries, employment has been essentially flat since late 2022. This represents a dramatic reversal from the robust growth these sectors experienced throughout the 2010s and early 2020s.

The data shows remarkable consistency across different analytical approaches. The researchers conducted multiple robustness checks:

• Excluding Covid era: Results remain statistically significant
• Dropping 2022-2023: The employment deceleration persists
• Alternative timing: Various ChatGPT launch date specifications yield similar findings

Importantly, coder employment remains close to the pre-Covid linear trend overall, but the study identifies a clear “kink” in the growth trajectory starting in late 2022. This suggests the effect is not simply a return to historical norms but represents a genuine structural break.

The timing aligns with external evidence from job market indicators. Indeed job postings data shows software developer openings fell more than 50% between 2022 and 2023, then stabilized and recently edged up slightly. This pattern mirrors the employment growth deceleration captured in the Federal Reserve analysis.

Geographic analysis reveals the trend is not concentrated in traditional tech hubs. The employment deceleration appears across metropolitan areas, suggesting this is a broad-based occupational shift rather than a Silicon Valley-specific phenomenon.

Industry Control Methodology: Separating AI from Industry Shocks

The Federal Reserve study’s most innovative methodological contribution is its industry control framework, which separates AI-specific effects from broader industry-level economic shocks. This addresses a crucial question: Are coders losing jobs because AI is substituting for their work, or because the industries employing them are declining?

The researchers developed a within-industry/between-industry decomposition that creates a counterfactual scenario showing what coder employment would look like if industry composition remained constant. This methodology is critical because industries were differentially substituting away from coders—an occupation-specific shock distinct from industry-level decline.

Post-ChatGPT Findings with Industry Controls (Table 6)

The consistency of the -3% annual effect across all industry categories is remarkable. It demonstrates that the deceleration is not attributable to coders working in declining industries. Instead, growing and stable industries are specifically reducing their coder hiring relative to other occupations.

The identifying assumption underlying this methodology is that industry-level shocks (changes in demand, total factor productivity, or regulatory environment) affect employment homothetically—they scale all occupations within an industry proportionally without changing occupational composition. When this assumption holds, deviations from proportional scaling indicate occupation-specific effects.

To validate this approach, the researchers conducted a placebo test using the lowest AI-exposure occupational quintile. This placebo group shows no significant post-GPT employment changes in most specifications, supporting the hypothesis that high AI exposure is driving the observed effects among coders.

The methodology’s power comes from its ability to separate multiple simultaneous effects:

• Industry demand shocks: Some sectors growing faster than others
• General economic conditions: Interest rates, inflation, post-Covid adjustment
• Occupation-specific substitution: AI replacing coding tasks specifically
• Measurement error: Survey sampling and seasonal adjustment issues

By controlling for the first three factors, the study isolates the AI substitution effect with unprecedented precision. This represents a significant methodological advance over previous studies that struggled to separate technology effects from confounding economic factors.

Scale of Impact: 500,000 Job Shortfall Analysis

The Federal Reserve study’s most headline-grabbing finding is its estimate that roughly 500,000 additional coder jobs would have existed in the absence of large-scale AI use. This calculation provides concrete scale for AI’s employment impact, but requires careful interpretation.

The calculation methodology is straightforward: starting with 5.735 million coders in November 2022, applying the observed 3% annual growth reduction over approximately three years yields: 5.735 million × (1 – 0.97³) ≈ 500,000 jobs.

This estimate aligns remarkably well with independent research. Brynjolfsson et al. (2025) found a 12% employment decline for 22-25 year olds in top AI-exposure occupations, which translates to approximately 475,000 jobs lost using their methodology—a crude but consistent approximation.

Supporting Evidence from Job Market Indicators

External labor market data corroborates the scale of impact:

• Indeed job postings: Software developer openings fell more than 50% between 2022-2023
• LinkedIn hiring metrics: Tech roles showing sustained lower demand relative to 2021-2022 peaks
• University career services: Computer science graduates reporting longer job search times
• Salary trends: Modest wage pressure in entry-level programming roles

However, the researchers emphasize crucial caveats about interpreting the 500,000 figure:

“This does not represent 500,000 unemployed individuals. Many displaced coders would find employment in other occupations, and AI may be altering task composition so ‘potential coders’ enter adjacent roles that now incorporate more coding skills.”

The study cannot account for general equilibrium effects that might offset direct substitution:

• Productivity gains: AI-enhanced coding enabling new products and services
• Cost reductions: Cheaper software development expanding market demand
• Adjacent job creation: New roles in AI model training, prompt engineering, and AI-human collaboration
• Task reorganization: Non-coders performing AI-assisted coding tasks

The timing pattern adds nuance to the 500,000 estimate. The gap between actual and counterfactual employment has widened significantly since mid-2024, suggesting acceleration rather than immediate impact following ChatGPT’s launch. This delayed response pattern indicates firms needed time to integrate AI tools into workflows before employment effects became measurable.

Industry-Specific Impacts Across Sectors

The Federal Reserve analysis reveals that AI’s impact on coder employment varies significantly across industries, with some sectors experiencing more dramatic shifts than others. A striking finding is that about 40% of U.S. coders work in computer systems design and related services (NAICS 5415)—not at high-profile Silicon Valley tech firms, but in contract software development.

Top Coder-Intensive Industries by Employment Concentration

The concentration in computer systems design is particularly significant because these firms provide contracted development services across the economy. When this sector reduces coder hiring, it affects software development capacity for businesses in all industries—from healthcare systems to manufacturing companies to financial services.

Industry-Level Employment Growth Patterns (CES Data)

The Information sector, which includes many tech companies, has fallen below its 2019 employment level. However, the researchers note this may reflect a return to trend after Covid-era overhiring rather than AI-specific effects alone.

The industry analysis reveals important nuances:

• Contract development most affected: NAICS 5415 (computer systems design) shows the most pronounced deceleration
• Software publishers stabilizing: Large tech companies have largely completed their post-Covid workforce adjustments
• Embedded coders resilient: Coders in manufacturing, healthcare, and financial services show less pronounced effects
• Startup ecosystem impact: Early-stage companies increasingly able to build products with smaller technical teams

Geographic patterns within industries also matter. The researchers find that the employment deceleration is not concentrated in traditional tech hubs like San Francisco or Seattle, but appears broadly across metropolitan areas. This suggests AI adoption is democratizing access to coding capabilities, reducing demand for location-specific technical talent clusters.

The industry-specific findings have important policy implications. If AI primarily affected high-wage tech workers in expensive coastal cities, the economic adjustment might be different than if it broadly impacts technical workers across the country in various industries.

AI Exposure Measurement: Why Metrics Disagree

One of the Federal Reserve study’s most important methodological insights concerns the measurement of AI exposure itself. The research reveals that the two leading AI exposure measures disagree about half the time on which occupations are most vulnerable to AI substitution.

The study compares two prominent metrics:

• Eloundou et al. (2024) “GPTs are GPTs”: Based on task analysis and expert judgment about AI capabilities
• Handa et al. (2025) Anthropic Economic Index (AEI): Based on actual Claude AI usage patterns by occupation

AI Exposure Agreement Matrix (% of Total Employment)

The striking findings from this comparison:

• Only ~10% of workers are classified as high exposure by both measures
• ~18.7% are classified as high by one measure but not the other
• Both agree coders are highly exposed (99.5% GPTs, 98.2% AEI)
• Disagreement is substantial for many other occupational categories

The disagreement has important implications for AI impact research and policy. If researchers cannot agree on which occupations are most exposed to AI, it becomes difficult to predict where employment effects will appear next.

Potential Explanations for the Disagreement

Timing differences: The AEI captures current AI use patterns, while the GPTs metric attempts to predict eventual exposure based on technical capabilities. AI adoption may be slower than technical potential suggests.

Data limitations: AEI data excludes API users and business licenses, potentially undercounting professional usage. ChatGPT usage data shows only 4.2% of prompts relate to programming (versus much higher proportions in Claude), possibly because ChatGPT serves more personal use cases.

Task granularity: The GPTs approach analyzes detailed task requirements, while AEI reflects aggregate occupational usage. Some occupations may use AI for specific tasks without overall high exposure.

Adoption barriers: Technical capability doesn’t guarantee adoption. Regulatory constraints, workflow integration challenges, and organizational inertia may slow AI uptake in theoretically high-exposure occupations.

For policymakers, this measurement disagreement suggests caution in making broad predictions about AI’s employment impacts beyond the most clearly exposed occupations like coding. The Federal Reserve study’s focus on coders partly reflects the unusual consensus between exposure metrics for this occupational group.

Historical Validation: Methodology Testing

The Federal Reserve researchers validated their industry decomposition methodology by testing it against known historical cases of occupation-specific technological change. This historical validation strengthens confidence that the ChatGPT-era findings reflect genuine AI substitution rather than methodological artifacts.

Successful Historical Validations

Bank Tellers and ATMs: The methodology correctly identifies large negative occupation-specific shocks for tellers despite massive banking industry growth during ATM deployment periods. This aligns with extensive literature on ATM-teller substitution.

Bookkeepers and Computerization: The analysis captures the well-documented decline in bookkeeping occupations following widespread computer adoption, correctly attributing it to occupation-specific rather than industry-level factors.

Telephone Operators: The methodology identifies the dramatic occupation-specific decline following automated switching systems, consistent with telecommunications industry growth during the same period.

Data Entry Keyers: The analysis correctly shows initial occupation-specific growth with computerization in the 1980s, followed by decline from internet adoption and offshoring in the 1990s.

Correct Attribution of Industry vs. Occupation Effects

Textile Sewing Machine Operators: Employment decline is correctly attributed primarily to industry-level shocks (international trade) rather than occupation-specific technological substitution.

Petroleum Engineers: Industry-level ups and downs from oil market fluctuations are correctly captured while occupation-specific effects remain minimal.

Legal Assistants and Paralegals: Positive occupation-specific shocks are correctly identified, reflecting increased demand for legal support services despite stable legal industry employment.

CNC Programmers: The methodology captures positive occupation-specific effects from manufacturing automation adoption.

This historical validation addresses a key concern about the study’s findings. Critics might argue that the observed coder employment patterns reflect economic factors unrelated to AI—post-Covid adjustment, interest rate increases, or other industry-specific shocks. The successful identification of known historical technology-labor substitution patterns using identical methodology provides strong support for the AI interpretation.

The validation also demonstrates the methodology’s ability to distinguish between different types of economic shocks:

• Technology substitution: Occupation-specific decline within growing or stable industries
• Industry decline: Proportional employment reduction across all occupations within affected industries
• Task reorganization: Changes in occupational composition reflecting new production methods
• Demand shifts: Industry growth or decline from external market forces

The historical evidence supports the interpretation that AI represents genuine technological substitution for coding tasks rather than a cyclical or industry-specific adjustment.

Confounding Factors and Threats to Analysis

The Federal Reserve study acknowledges several alternative explanations for the observed employment patterns, demonstrating scientific rigor in considering threats to causal identification. The most significant potential confounding factor is the Tax Cuts and Jobs Act (TCJA) provision that took effect in 2022.

Tax Law Changes and R&D Investment

TCJA R&D Provision: Starting in 2022, research and development expenses had to be amortized over five years instead of being immediately expensed, effectively raising the cost of software development activities.

The Information sector accounts for approximately 25% of total U.S. R&D spending, making this policy change particularly relevant for tech companies. However, empirical evidence on the TCJA’s actual impact remains mixed:

• Cowx et al. (2025): Find evidence of reduced R&D spending following the policy change
• Du and Li (2025): Find no significant reduction in R&D investment
• Industry surveys: Suggest delayed rather than cancelled development projects

Encouragingly for the AI interpretation, the coder employment effects survive in non-coder-intensive industries where TCJA impacts would be less salient. This suggests tax policy changes alone cannot explain the occupation-specific employment patterns.

Other 2022 Economic Factors

Interest Rate Increases: The Federal Reserve began raising rates in March 2022, which could have disproportionately affected growth-oriented tech companies. However, the employment deceleration timing aligns more closely with ChatGPT’s November 2022 launch than with initial rate increases.

Post-Covid Reopening Effects: Consumer spending patterns shifted from online services toward in-person activities, affecting digital advertising revenue at Google and Meta in 2022. This industry-specific demand shift might explain some employment patterns, but wouldn’t account for the cross-industry occupation-specific effects the study documents.

Cryptocurrency Market Collapse: The crypto winter and FTX collapse in November 2022 eliminated significant demand for blockchain developers. However, crypto-specific employment represents a small fraction of total coder employment.

Labor Market Dynamics and Measurement Issues

Wage Effects Absence: The study finds no visible break in average coder wages in 2022, suggesting the main effect operates through employment quantities rather than prices. This pattern is consistent with firms having some monopsony power or with composition effects masking wage impacts.

Anticipated vs. Realized Productivity: Firms may be reducing hiring in anticipation of AI substitution even before experiencing actual productivity gains. Economic theory suggests a “band of inaction” with fixed hiring and firing costs, where employers respond to expected rather than realized changes.

Survey Measurement Challenges: Current Population Survey sampling may not perfectly capture rapid changes in employment patterns, particularly for smaller occupational categories. Seasonal adjustment procedures could potentially create spurious breaks in time series data.

The researchers address these concerns through multiple robustness checks, alternative timing specifications, and placebo tests using low-AI-exposure occupations. The consistency of results across different analytical approaches strengthens confidence in the AI interpretation while acknowledging that multiple factors likely contribute to observed employment patterns.

What Research Cannot Address: Open Questions

The Federal Reserve study represents the first rigorous quantitative analysis of AI’s employment effects, but the researchers are careful to acknowledge significant limitations and open questions that their methodology cannot resolve. These limitations are crucial for interpreting the findings’ implications for policy and business strategy.

Demand Elasticity and Long-Term Effects

Elastic Demand Possibility: If demand for coding services is elastic, cheaper AI-assisted development could eventually increase coder employment by enabling new products and markets. The study captures only short-term substitution effects, not potential long-term demand expansion.

Historical parallels suggest caution in predicting long-term outcomes. Previous automation waves often led to short-term displacement followed by employment recovery as new applications emerged. The spreadsheet software example illustrates this pattern—initially reducing demand for bookkeepers but eventually expanding financial analysis roles across many industries.

New Work Creation and Task Reorganization

Acemoglu-Restrepo Framework: The study cannot capture “new work” creation—entirely new tasks and occupations that emerge from AI capabilities. Large language models may enable new coding-intensive products that don’t exist today, potentially creating employment categories not captured in current occupational classifications.

Task Reorganization: AI might enable non-coders to perform coding-adjacent tasks with AI assistance, effectively expanding the pool of workers who can contribute to software development. This could increase total coding activity while reducing demand for traditional coding roles.

General Equilibrium Effects

The researchers emphasize that automation typically raises aggregate productivity and labor demand for all workers: “the average worker is better off after a positive productivity shock.” However, measuring these economy-wide effects requires longer time horizons and more complex modeling approaches.

Key general equilibrium channels the study cannot address:

• Productivity spillovers: AI-enhanced coding reducing costs across all industries
• Consumer surplus effects: Better software improving living standards
• Capital-labor complementarity: AI tools requiring more skilled human oversight
• Innovation acceleration: Faster product development cycles creating new market opportunities

AI Industry Employment Creation

Direct AI Employment: Companies like OpenAI, Anthropic, and Google DeepMind likely employ fewer than 15,000 people combined. Even multiplying by 6x for the broader AI industry yields less than 2% of U.S. coders, insufficient to offset the documented employment deceleration.

However, this calculation may underestimate AI-related job creation in established companies integrating AI capabilities, consulting firms helping businesses adopt AI, and entirely new categories of AI-human collaboration roles.

Geographic and Remote Work Implications

Offshore and Contract Work Vulnerability: The study cannot determine whether remote and offshore coding services providers experience more severe employment impacts than domestic full-time employees. If AI reduces demand for routine coding tasks disproportionately, it might particularly affect distributed development models.

Geographic Rebalancing: AI might reduce the premium for coding talent in expensive tech hubs if enhanced productivity makes location less important for software development.

Broader Automation Potential

LLM-Using Coders as Automation Agents: The study focuses on AI substituting for coders but cannot measure coders using AI to automate other business processes. This secondary effect might have larger long-term employment implications than direct coder substitution.

Cross-Occupational Spillovers: Successful AI adoption in coding might accelerate automation efforts in adjacent technical fields like data analysis, system administration, and technical writing.

Policy Implications and Forward-Looking Assessment

The Federal Reserve study provides policymakers with the first concrete evidence that artificial intelligence is having measurable economic impacts on employment, moving AI policy discussions from theoretical frameworks to empirical realities. The findings carry significant implications for workforce development, education policy, and economic regulation.

Key Policy Takeaways

1. AI is Having Measurable Employment Effects
The approximately 3% annual growth reduction for coders represents the first rigorous quantitative evidence of AI’s labor market impact. This finding validates concerns about technology-driven job displacement while providing a concrete scale for policy response.

2. Effects Are Occupation-Specific, Not Industry-Driven
The fact that industries are substituting away from coders even while growing suggests targeted rather than broad-based economic adjustment needs. Policies should focus on occupational retraining rather than regional economic development.

3. Employment Continues Growing, But Slowly
Coder employment has not collapsed—it continues growing at roughly 1% annually versus 4.9% pre-ChatGPT. This suggests gradual rather than sudden labor market adjustment, providing time for policy response.

4. Effects Appear to Be Accelerating
The widening gap between actual and counterfactual employment starting mid-2024 suggests AI adoption impacts are accelerating rather than plateauing. This trend demands proactive rather than reactive policy approaches.

Workforce Development Implications

Educational Curriculum Adaptation: Computer science programs may need to emphasize AI-human collaboration skills, system design, and domain expertise over routine programming tasks. The study suggests demand is shifting toward roles that effectively combine AI tools with human judgment.

Retraining Program Design: The estimated 500,000 job shortfall (absorbed into other occupations) suggests significant need for career transition support. Programs should focus on transferable skills like logical reasoning, problem-solving, and technical communication that remain valuable in AI-augmented work environments.

Professional Certification Evolution: Industry certifications may need to incorporate AI tool proficiency alongside traditional technical skills, reflecting how professional competency definitions are evolving.

Economic Policy Considerations

Unemployment Insurance Adaptation: Traditional unemployment systems assume temporary layoffs with eventual return to similar occupations. AI-driven displacement may require longer retraining periods and career transition support extending beyond current benefit durations.

Tax Policy Coordination: The interaction between the TCJA R&D provisions and AI adoption suggests need for coordinated policy analysis. Tax incentives for R&D might need adjustment to account for AI’s productivity effects on development costs.

Antitrust and Market Concentration: If AI advantages concentrate among large technology companies with the resources to develop and deploy advanced models, employment effects might be accompanied by increased market concentration requiring regulatory attention.

Timeline and Forward-Looking Assessment

Historical Timeline (Established by Study):

• Pre-ChatGPT (2016-2022): ~4.9% annual coder employment growth
• November 2022: ChatGPT release marking AI mainstream adoption
• 2022-2023: Initial slowdown, software job openings fall >50%
• Mid-2024: Employment gap widens significantly
• 2024-2025: Job openings stabilize, employment growth remains flat
• March 2026: Cumulative estimated shortfall reaches ~500,000 jobs

Forward-Looking Scenarios:

Optimistic Scenario: Demand elasticity leads to employment recovery as AI-enabled products create new market categories. Historical precedent suggests technology that reduces production costs eventually expands markets and employment, but with different skill requirements.

Continued Deceleration Scenario: AI capabilities continue advancing faster than new demand creation, leading to sustained slower employment growth. This scenario suggests need for proactive workforce transition policies.

Acceleration Scenario: AI advances to automate higher-level programming tasks, extending employment impacts to senior developers and system architects currently less affected. This would require more comprehensive policy responses.

International Competitiveness Considerations

The study’s findings have implications for U.S. technology competitiveness. If AI reduces the comparative advantage of having large pools of skilled programmers, it might level the playing field between countries with different technical education capacities. Alternatively, countries that most effectively integrate AI into their development processes might gain competitive advantages.

Policy responses should consider how workforce adaptation strategies affect national technological leadership while supporting displaced workers’ economic security and career development opportunities.

The Federal Reserve study marks a turning point in AI policy discussions—moving from speculation about potential impacts to evidence-based analysis of measurable effects. The challenge for policymakers is developing responsive frameworks while AI capabilities and adoption patterns continue evolving rapidly.