Lore

One-Shotting as a Workflow Goal

Overview

One-shotting is the practice of aiming to get a complete, correct result from an AI agent in a single pass, rather than iterating with it live in a back-and-forth chat. The success metric shifts from "did I guide it to a good result" to "did the first attempt land," which changes where effort goes: instead of steering mid-generation, you front-load effort into the spec or prompt.

Mechanism

Because the agent won't be corrected turn-by-turn, the upfront brief has to carry everything the agent would otherwise learn from clarifying questions or live feedback — constraints, acceptance criteria, examples. This is the practical payoff of Specification-Driven Development: a strong spec is what makes one-shotting viable instead of producing plausible-but-wrong output.

Apply

Write the extra detail into a prompt or spec before launching an agent task, specifically so you can walk away and let it run rather than babysitting the session. Reserve live, turn-by-turn steering (see interrupt-and-steer-agent-technique) for cases where one-shotting predictably fails.

Attribution

Ryan Nystrom (Notion), describing his personal Codex workflow — practitioner anecdote, not a measured result.