Knowledge work distribution before and after AI

A new report from the American Enterprise Institute focused on the U.S. economy sets out that AI is creating an "up or out" reality for knowledge workers, mirroring the manufacturing deindustrialisation of previous decades. The research examines four major sectors –finance, business services, government, and healthcare– finding that mid-level workers face the greatest vulnerability as AI automates routine cognitive tasks.

Shift in Workforce Talent Distribution: Before and After AI

While technical coding jobs decline (software developer employment has dropped 15% since 2022), demand surges for workers who combine AI fluency with emotional intelligence, critical thinking and adaptability. For example, the Federal Reserve Bank of New York reported in 2023 that unemployment rates for liberal arts graduates were half that of computer scientists (Note: this does not indicate the type of employment liberal arts graduates secured. This Entrepreneur article gives further insights, including that median early-career earnings for computer scientists is almost double that of arts/humanities graduates).

US Software Developer Employment Index

Key points:

• 30% of US workers could see at least half their tasks disrupted by generative AI, particularly in administrative support and data entry roles
• The adoption of AI systems in professions undertaking knowledge work can lead to the "compression effect": middle-skill level workers face pressure to either upskill to supervise AI systems or move to lower-skilled positions
• Social skills premium: Jobs requiring high social skills but low math skills show significant growth, while high math/low social roles decline

The report warns against repeating past policy failures. During manufacturing automation, inadequate worker support created lasting economic and social costs. Policy recommendations include proactive interventions prioritising worker agency and lifelong learning:

• Individual Training Accounts: Personal funding accounts that let workers choose their own retraining programmes instead of being assigned by government agencies
• Automation Adjustment Assistance: Comprehensive support for workers displaced by AI, including wage insurance, retraining funds, and relocation help (modelled on past trade displacement programmes)
• Universal AI literacy: Teaching AI skills from primary school through executive training, integrating AI as a basic skill alongside reading and writing
• Early childhood investment: Strengthening programmes that develop emotional and social skills, which become more valuable as AI handles technical tasks

Source: Orrell, B. 2025. “De-Skilling the Knowledge Economy.” Research Paper, June 2025. American Enterprise Institute.