---
schema: 1
type: term
title: '**Thinking box**'
description: >-
  **Thinking box** — in this channel's material, a place in a knowledge-software
  product where AI is given space to reason on the user's behalf; Notion
  cofounder Simon Last describes AI **“thinking boxes”** as combining with
  Notion's core primitives (databases, UI, and permissions) to build
  LLM-friendly interfaces.
canonical: 'https://liquidgold.cc/channel/everyinc/explore/term/thinking-box-7b3246646ba1'
markdown: >-
  https://liquidgold.cc/channel/everyinc/explore/term/thinking-box-7b3246646ba1.md
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# \*\*Thinking box\*\*

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/thinking-box-7b3246646ba1

Parent: [Channel index](https://liquidgold.cc/channel/everyinc/llms.txt)

## Definition

**Thinking box** — in this channel's material, a place in a knowledge-software product where AI is given space to reason on the user's behalf; Notion cofounder Simon Last describes AI **“thinking boxes”** as combining with Notion's core primitives (databases, UI, and permissions) to build LLM-friendly interfaces.

In this channel's material, **thinking boxes** show up when Notion cofounder **Simon Last** describes how the building blocks of knowledge software — *databases*, *UI*, and *permissions* — combine with AI. The term refers to the places in a knowledge product where the model gets room to work through a problem on the user's behalf, woven into the interface rather than bolted on as a chat window.The framing connects to a broader point from the same conversation: designing an **LLM-friendly interface** is central to shipping AI software at scale, because stochastic behavior needs evaluation and surfaces that make the model's work legible to users. Thinking boxes are one way those surfaces manifest in a real product serving hundreds of millions of users.- **Thinking boxes** are Notion's pattern for embedding AI reasoning inside its existing knowledge-software primitives (databases, UI, permissions).