OpenAI reaches its ‘automated research intern’ goal

OpenAI says it has achieved the “automated research intern” goal it set in fall 2025. The milestone describes a supervised system for well-defined research tasks, not an independent scientist. Under human direction, it can carry out work that would take a skilled researcher a few days. A fuller automated AI researcher remains a target for March 2028.

Three agent-workdays inside one human workday

As of mid-August 2026, OpenAI’s research organization used 3.1 agent-workdays of effort for every workday of human labor, with each workday defined as eight hours. That figure measures machine runtime and effort, including work performed in parallel; it is not evidence of a 3.1-times productivity gain.

The median researcher was using more than $600 per day of coding-agent inference at API prices, while the 90th-percentile user exceeded $7,000. These are API-price equivalents rather than a statement of OpenAI’s internal marginal costs. Experiments per active experimenter also reached their highest level since tracking began in January 2025, correlated with wider Codex adoption as available compute grew significantly.

High-level planning remains rare in agent output

OpenAI grouped agent work into Decide, Design, Build, Run, Analyze, and Communicate. Every category increased between January and August 2026, but high-level planning still represented only a minimal fraction of agent output tokens. Over the past six months, more than half of successful tasks estimated to require four to eight hours of human work involved at least one human intervention. People continue to set research priorities, choose which ideas and results to pursue, and decide whether systems should be scaled, paused, or deployed.

Acceleration arrives with a warning on safety

OpenAI cautions that research has many bottlenecks, so overall progress is unlikely to keep pace with raw increases in agent usage, code, or experiments. In a separate post published the same day, chief scientist Jakub Pachocki called for extreme caution and said he believes no lab has solved alignment and monitoring well enough to keep scaling responsibly at maximum speed for much longer. OpenAI says it does not yet know how to safely achieve aligned full recursive self-improvement and will slow or stop development or deployment when it encounters unacceptable risks that cannot be sufficiently safeguarded.