Drew Johnston. Researcher at OpenAI studying how agentic AI tools change knowledge work. Johnston and co-author David Holtz conducted the large-scale Codex deployment study covering OpenAI's own workforce across engineering, legal, research, and general knowledge roles from late 2024 through 2026. The study measured how skills-sharing mechanisms drive cross-functional AI adoption beyond the engineering teams that first deploy agentic tools.
Operating themes
- Agentic diffusion mechanics How agentic tools spread from engineering into legal, research, and knowledge-work roles when low-friction adoption mechanisms exist.
- Skills as an adoption lever Shareable workflow templates as the key variable separating roles that scale rapidly from those that plateau at early adoption.
- Depth over breadth Measuring AI impact through output complexity growth (8-hour task volume, tokens per user) rather than user count alone.