Agent Workday Index (AWD)
The Agent Workday Index (AWD) measures how many workdays an AI agent delivers per human workday. OpenAI introduced the metric with its Automated Research Intern program: as of September 2026, the index stands at 3.1 — every human workday produces the equivalent of 3.1 agent workdays. The AWD is a multiplier, not an absolute hour count, which makes it comparable across teams. Combined with daily spend, it yields the cost per generated agent workday: at a median daily spend of $600, that is roughly $194 per agent day; for top-10-percent users above $7,000, about $2,258. OpenAI targets an index of 6.0 by March 2028. The key limit: the index counts delivered output days, not their quality per task, so teams should pair it with their own ticket throughput. It sits alongside related concepts such as Agent Economics and Agentic Compute as the most practical metric for running coding and research agents economically.
Deep Dive: Agent Workday Index (AWD)
The Agent Workday Index (AWD) measures how many workdays an AI agent delivers per human workday. OpenAI introduced the metric with its Automated Research Intern program: as of September 2026, the index stands at 3.1 — every human workday produces the equivalent of 3.1 agent workdays. The AWD is a multiplier, not an absolute hour count, which makes it comparable across teams. Combined with daily spend, it yields the cost per generated agent workday: at a median daily spend of $600, that is roughly $194 per agent day; for top-10-percent users above $7,000, about $2,258. OpenAI targets an index of 6.0 by March 2028. The key limit: the index counts delivered output days, not their quality per task, so teams should pair it with their own ticket throughput. It sits alongside related concepts such as Agent Economics and Agentic Compute as the most practical metric for running coding and research agents economically.
Implementation Details
- Tech Stackopenai
- Production-Ready Guardrails