When the Code Writers Are Automated: An Agent-Based Simulation of the Macroeconomic, Distributional and Financial Effects of AI-Driven Software Job Displacement

If artificial intelligence (AI) removes a large share of jobs in the software industry, what happens to the rest of the economy? Software workers are a small share of employment, but they are high earners with large savings, mortgages and consumption footprints, and their spending supports shops, services and manufacturing. We build a stylised agent-based model of a closed-loop economy with 10,000 worker-households, a capital-owning class, six sectors, 800 individually modelled retail shops, a mortgage-lending banking system and a government with progressive taxes and unemployment insurance. We then simulate AI displacing 30%, 60% and 90% of the labour required per unit of IT output over three years, and track the economy for ten years across 20 random seeds per scenario. In the central 60% scenario, unemployment is 14.7 percentage points higher than in a no-AI counterfactual after ten years (18.8% versus 4.1%), household consumption is 14.1% lower, gross output is 5.3% lower, the wage share of output falls from 61% to 48%, the income Gini coefficient rises from 0.50 to 0.58, mortgage defaults reach 20.7% of mortgages (6.4% in the baseline), and about 3% of shops close. A policy package of longer unemployment insurance, reskilling and a windfall tax on IT profits halves mortgage defaults and reduces the unemployment increase by about a quarter, but does not prevent it. The results are most sensitive to the speed at which the labour market re-employs displaced workers, the strength of the credit channel, and the extent to which AI profits are recycled into domestic demand. The model is stylised and not calibrated to a specific country; its results should be read as mechanisms and orders of magnitude, not forecasts. Code and data: https://github.com/tejkumar7/ai-job-displacement-abm