Is AI Doom Going Out of Style?

What It Covers

A potential narrative shift away from AI job doom is emerging simultaneously in economic commentary and financial markets. Ezra Klein's New York Times piece, software engineering job data showing 18% growth, Anthropic's ARR reportedly hitting $44 billion, and Stripe Atlas incorporations up 130% year-over-year signal a more nuanced conversation replacing binary doom framing.

Key Questions Answered

• •Jevons Paradox applied to AI labor: When AI reduces the cost of a task, total demand for that work often expands rather than contracts. Software engineering job postings are up 18% since May 2024, and Federal Reserve data shows software roles at their highest levels since November 2023, directly contradicting displacement narratives with measurable labor market evidence.
• •Elastic vs. inelastic demand framework: Not all work responds equally to AI cost reductions. Elastic domains — software development, legal discovery, sales outreach, security monitoring — expand when costs fall. Inelastic domains — payroll, compliance filing, routine reporting — have capped demand. Identifying which category your work falls into determines whether AI creates or eliminates roles in your sector.
• •Tokens replace seats as the AI revenue model: The shift from per-seat SaaS pricing to token-based consumption removes the ceiling on AI revenue. A single developer using Claude Code or Codex can generate hundreds to thousands of dollars monthly in token spend versus a flat $20 subscription, explaining why Anthropic's ARR reportedly doubled from $9 billion to $44 billion within months of 2025.
• •Structured knowledge graphs outperform RAG for enterprise AI: Atlassian's Rovo tool demonstrates that platforms with 20 years of structured relational data — linking teams, code, people, and work — can answer agent queries via graph lookups rather than token-heavy retrieval-augmented generation. Customers using Rovo grew their own ARR at twice the pace of non-users, and Atlassian stock rose nearly 30% on earnings.
• •Partial displacement is harder to address than mass unemployment: Economist-aligned analysis suggests AI is more likely to eliminate specific task categories within jobs than entire occupations wholesale. Historical China trade competition displaced roughly 2 million U.S. workers — statistically small against 5 million monthly hires — yet caused severe localized harm with minimal policy response, a pattern worth anticipating for AI-driven displacement.