---
title: "Knowledge"
summary: "Concepts and Guides on engineering with and for AI agents, by category."
published: 2026-11-09
updated: 2026-11-09
language: en
tags: []
canonical: https://joaodavidsilva.com/knowledge/
---

# Knowledge

## AI Engineering

- [Agent](https://joaodavidsilva.com/knowledge/agent.md): A model harnessed with tools, a system prompt and a context window that takes turns with a user: the unit you actually talk to.
- [Model](https://joaodavidsilva.com/knowledge/model.md): The parameters: a stateless next-token predictor that can't do anything agentic until it is harnessed.
- [Token](https://joaodavidsilva.com/knowledge/token.md): The atomic unit a model reads and writes; context window size, cost and latency are all counted in tokens.

## Context Engineering

- [Clearing](https://joaodavidsilva.com/knowledge/clearing.md): Ending the current session and starting a fresh one with an empty context window.
- [Compaction](https://joaodavidsilva.com/knowledge/compaction.md): A handoff done in-memory: the session's history is summarised and the summary seeds a fresh session.
- [Context](https://joaodavidsilva.com/knowledge/context.md): The task-relevant information the agent has access to right now, as distinct from the context window and the session.
- [Context window](https://joaodavidsilva.com/knowledge/context-window.md): Everything the model sees on each model provider request: finite, model-specific, and the only surface through which it perceives anything.
- [Session](https://joaodavidsilva.com/knowledge/session.md): One bounded run of interaction with an agent; the session is what fills the context window.
- [Smart zone](https://joaodavidsilva.com/knowledge/smart-zone.md): The early, sharp part of a session, before growing context drifts the agent into a sloppier "dumb zone".

## Harness Engineering

- [Harness](https://joaodavidsilva.com/knowledge/harness.md): Everything around the model that turns it into an agent: tools, system prompt, context-window management, permissions, hooks.
- [Turn](https://joaodavidsilva.com/knowledge/turn.md): One user message plus everything the agent does in response, up until it yields back to the user.

## MCP & Tooling

- [Tool](https://joaodavidsilva.com/knowledge/tool.md): A function the harness exposes for the agent to call, and the only way the agent perceives and acts on its environment.
- [Tool call](https://joaodavidsilva.com/knowledge/tool-call.md): The model's output naming a tool and its arguments: structured text that the harness has to read and execute.