Ken Thompson added pipes to Unix in 1973, in Version 3. A pipeline chains programs so the output of one becomes the input of the next.
A shell pipeline is fixed by the person who types it. An agent composes its own at run time, because the model reads the result of one tool call and decides which call comes next.
A single input can therefore take a different path on the next run, and the code around the model has to expect that and recover from it.
The engineering in an agent sits outside the model, in the harness that runs the loop and the handling for when it goes wrong.
What an agent is
- Types of AI Agents: Five Types by Autonomy and Scope. Groups agents into five types by autonomy and scope, to decide how much autonomy a task needs.
- AI Agent vs AI Assistant: When to Build Each. Draws the line at who chooses the next step, with an experiment in which six assistants were told to act alone.
- AI Agent Design Patterns. Sets Anthropic's six agent design patterns beside a second list and gives a grid for placing any pattern.
The loop and the harness
- The Anatomy of an Agent Loop: Perceive, Think, Act, Remember. Breaks the loop into four phases, which helps locate the phase where a run went wrong.
- Agent Harnesses: The Infrastructure Your LLM Agent Needs. Lists what a harness contains and explains why a framework alone may not supply it.
- Lambda Calculus as an AI Reasoning Exercise. Uses lambda calculus problems with mechanically checkable answers to expose reasoning errors in model output.
Protocols
- Model Context Protocol (MCP): Connecting AI Tools. Explains how MCP connects a model to data sources and tools.
- MCP Server Tutorial: Build a Docs Search Server in Python. Builds a docs search server in Python and tests it without a model.
A single agent or several
- Multi-Agent vs Single-Agent Systems: The Real Trade-offs. Weighs one agent against several on failure modes and latency.
Failure
- Error Handling in Production AI Agents. Covers the error patterns that break agents in production and the handling that contains them.
- Why AI Agents Fail in Production. Catalogs why agents fail in production and what the field is doing about it.
How do the parts of an AI agent fit together?
The model chooses the next tool call, the loop carries the result back to it, and the harness runs the loop. The agent loop page covers the four phases, and the agent harness page covers the layer around them.
Related topics
The loop's remember step has a topic of its own, and AI agent memory covers what gets written and read back.
A coding agent is the agent a developer uses daily, and the coding agents topic covers the files that configure one.
Each turn of the loop spends tokens and latency, which the LLM inference topic measures.