Why AI Job Losses Are Both Real and Overstated

James Wang runs the numbers on AI-driven productivity and lands somewhere neither doomers nor dismissers want to be. MIT field experiments across Microsoft, Accenture, and a Fortune 100 company showed coding assistants boosted task completion by about 26%, but once you factor in adoption rates, bug rates in AI-generated code, and the fact that actual coding is only 11-32% of developer time, realistic project-level gains shrink to roughly 10%. That still translates to around $36 billion in added output from software alone - already exceeding Daron Acemoglu's estimate of total AI economic contribution. Wang's framing tracks historical parallels like photography killing photorealistic painting while expanding visual work overall: mechanical skills get commoditized, judgment-based work commands a premium, and total employment in affected fields tends to grow rather than collapse.