AFK (Away From Keyboard)

AFK, or away from keyboard, describes a working pattern where a person starts an AI coding agent's session and then leaves it to run unattended, instead of staying to answer questions as they come up. Matt Pocock's AI Coding Dictionary (aicodingdictionary.com), a glossary of agentic-coding terms open-sourced on GitHub, calls it the throughput multiplier of AI coding: several AFK sessions can run in parallel while the person sleeps, eats, or does other work. Running AFK safely usually needs a permissive permission mode paired with sandboxing, since nobody is present to catch a dangerous action before it executes.

The dictionary's core observation is about ambiguity. While a person watches a session, an unclear decision surfaces as a question they can answer right away. Once they walk away, the agent instead picks a default and keeps building on that guess through every later decision. The characteristic failure is returning to hours of confident, internally consistent work built on a wrong call made in the first few minutes. The output isn't sloppy. It's coherent about the wrong thing.

Because no input can reach the agent mid-run, the dictionary recommends giving it instructions before and after instead: resolve ambiguity up front with a written spec, let automated checks and automated review substitute for the missing attention during the run, and end in something reviewable, a pull request, rather than code already merged. AFK doesn't remove human review. It defers all of it to the end. The dictionary prefers "AFK" to "background agent," since the latter describes the machine while AFK names the fact that matters: nobody is watching.