# OpenAI's Decisions API answers only from choices you write. Put one in front of every agent action.

*2026-10-27 ⬩ 2:30 PM ET | bryn*

OpenAI made its Decisions API generally available on October 22, and it's built for the small call an agent makes right before it acts. Per [OpenAI's October 22 post](https://openai.com/index/decisions-api/), you write a question and its allowed answers, and it returns a probability, a choice or a score.

One of OpenAI's own sample choices is whether an agent should continue, retry, or ask for clarification. Datadog uses it to score the risk of shell commands its agent wrote, before running them. Sierra uses it to pick the next action in live customer conversations, in under 280 ms at p95.

Here's what I'd do. Find the step where your agent decides to act, and stop letting it answer in prose. Write the answers it may give: act, wait, ask a person. A paragraph can argue itself into anything. A fixed menu can't add an option you never wrote, and every answer is one you can count later.

At $0.10 per million input tokens and no output charge, the check is cheap. Skipping it isn't.

## Sources

1. [Build more responsive apps with the Decisions API](https://openai.com/index/decisions-api/). OpenAI, Oct 22, 2026.

See also: https://www.civic.com/field-notes/acting-got-easy-deciding-didnt

Source: https://www.civic.com/bin/decisions-api-fixed-answers