Field diffusion · Computer science

OpenAI measures coding-agent adoption across its research organization

Internal telemetry reports increased coding-agent use, code contribution and experiment activity across an AI research organization during 2026.

Summary

OpenAI reports that coding-agent use reached 3.1 agent-workdays per human workday by mid-August, using an eight-hour day convention. Code contribution and experiments per active experimenter increased alongside adoption. The report also describes human interventions in longer tasks and explains limitations of its internal activity measures.

AI role

Coding agents assist research implementation, infrastructure troubleshooting, experiment analysis and other researcher-directed tasks.

Narrative role

Observed use and workflow activity provide institution-level diffusion evidence beyond access offers, while retaining the distinction between measured activity and scientific progress.

Caveat

These are internal observational measures at one well-resourced lab. Growing compute, changing task mix and incomplete coverage confound interpretation; agent runtime is not equivalent to human labor or a causal productivity multiplier.