OpenAI data shows the 'capability overhang': legal and recruiting lag engineering in agentic AI adoption
Exp_Mark · x · 2026-11-17
A blog post by micro1 chief economist Mark Esposito and Liu Zhang, building on OpenAI's paper 'The Shift to Agentic AI: Evidence from Codex,' examines the 'capability overhang': most people confine AI to familiar tasks like writing, leaving advanced agent capabilities untapped.
Key points:
- Inside OpenAI, nearly all AI use became agentic across departments (engineering to legal, recruiting) in about two months;
- At other companies, engineering and data science led a similarly fast shift, but legal and recruiting roles saw only 2% agentic AI use as of June 2026;
- Same models, same product, radically different adoption — pointing to an organizational, not capability, gap.
The authors argue AI's economic impact depends on how widely these capabilities enter everyday work, requiring redesigned workflows, better delegation, and new productivity measures.
More from AGI Musings
- Use AI to build a longitudinal view of media narratives, dev suggests — traviscline · 2026-11-17
- Who owns the AI workflow when automation silently breaks? — djgoel123 · 2026-11-17
- Stanford essay revisits Scholem's Golem warning to reframe human-AI relationships — begusgasper · 2026-11-17
- Edgar Pavlovsky: '2x the time' no longer equals 2x productivity in the AI era — edgarpavlovsky · 2026-11-17
- Judea Pearl praises a one-slide explainer of how structural causal models turn causation into math — yudapearl · 2026-11-17
- Former OpenAI researcher Kuprel: who wins the AI race decides between moonshot optimism and total control — Kuprel · 2026-11-17