# Smart zone

Concept · [Building with AI](https://ecosystem.samcarlton.com/ai)

Also called: dumb zone, smart zone and dumb zone, smart zone vs dumb zone

A name for the early part of a [session](https://ecosystem.samcarlton.com/session), when an [agent](https://ecosystem.samcarlton.com/ai-agent) is focused and accurate. As the session fills up it drifts into a 'dumb zone' where it gets careless and forgetful, even though the model has not changed. The usual fix is to clear or [compact](https://ecosystem.samcarlton.com/compaction) and restart with a short plan.

**For example:** Early on the [agent](https://ecosystem.samcarlton.com/ai-agent) writes careful code; by late afternoon it is renaming things it was told to leave alone. The [founder](https://ecosystem.samcarlton.com/founder) saves a short note, clears the [session](https://ecosystem.samcarlton.com/session), and the quality comes back.

## Connects to (8)

- [Attention degradation](https://ecosystem.samcarlton.com/attention-degradation): The drop in an agent's sharpness as a session gets long.
- [Session](https://ecosystem.samcarlton.com/session): One stretch of work with an AI agent, from a blank start until it is cleared, closed or compacted.
- [Clearing](https://ecosystem.samcarlton.com/clearing): Ending a session and starting a completely empty one.
- [Compaction](https://ecosystem.samcarlton.com/compaction): Shrinking a long session by replacing its history with a summary and carrying on from that summary in a fresh context window.
- [Handoff](https://ecosystem.samcarlton.com/handoff): Passing work from one session to the next, or from one agent to another, so the new one can continue without the full history.
- [Context window](https://ecosystem.samcarlton.com/context-window): How much text an LLM can take into account at one time.
- [Hallucination](https://ecosystem.samcarlton.com/hallucination): When an AI model states something false or invented as though it were true.
- [Attention budget](https://ecosystem.samcarlton.com/attention-budget): A way of picturing attention as a fixed amount each token can spread across everything else in the input.

Map view: https://ecosystem.samcarlton.com/smart-zone