Ploba logo

Find the right AI tool for the job: apps, agents, APIs, MCP servers and datasets, explained plainly.

Get the AI Digest

Discover

  • Explore
  • Top Charts
  • Categories
  • Agents
  • MCP servers
  • Developer tools

Learn & community

  • Learn AI
  • AI Glossary
  • Blog
  • Hackathons
  • People
  • AI Digest

For makers

  • List your tool
  • Docs
  • Claim a listing
  • Advertise
  • Write for us

Company

  • About
  • Careers
  • Changelog
  • Trust
  • Security
  • Contact

© 2026 Ploba. All rights reserved.

  • Privacy
  • Terms
  • Cookies
  • Advertising
  • All policies
llms.txt
AI glossary
Agents

Agent Harness

Also called Harness, AI harness, LLM harness, Agent scaffolding

Everything around an AI model that turns it into a working agent: its instructions, tools, context handling, memory, sandbox, checks and permissions.

What is an agent harness?

An agent harness is the software wrapped around an AI model that lets it work as an AI agent. The model reads and writes text; the harness does everything else. It gives the model its instructions, runs the tools it asks for, decides what goes into its context window, keeps track of progress between sessions, and stops it from doing things it shouldn't.

A common shorthand is agent = model + harness. Coding agents such as Claude Code and Codex are examples: the same model can behave very differently depending on the harness it runs in.

What does a harness include?

  • Instructions: the system prompt and project files the agent reads at the start, such as an AGENTS.md file.
  • Tools: the actions the agent can take, like running commands, editing files, searching or browsing, often through the Model Context Protocol.
  • Context management: trimming, summarising or moving large outputs out of the context window so the agent doesn't lose track.
  • State and memory: progress notes, task lists and version history that carry work from one session to the next.
  • A sandbox: an isolated place where the agent's commands run safely.
  • Feedback: tests, linters and other checks that tell the agent when it got something wrong.
  • Guardrails and approvals: permissions, blocked actions and points where a person signs off (human in the loop).

What is it commonly confused with?

  • Harness, the company. Harness is also the name of a software delivery (CI/CD) platform. "Agent harness" has nothing to do with it.
  • The model itself. Swapping in a better model doesn't fix a weak harness, and a good harness can make an average model far more reliable.
  • An agent framework. A framework is a library for building agents; a harness is the specific setup one agent runs in. Frameworks are often used to build harnesses.
  • Agent orchestration. Orchestration coordinates several agents; a harness is what surrounds each one.

Why does it matter?

Most failures of AI agents in real work come from the harness, not the model: missing context, no way to check results, or nothing that carries progress forward. Improving the harness by observing failures and fixing what caused them is now called harness engineering. Read more in What is harness engineering?.

Category
Agents
Last reviewed
Nov 3, 2026
Also called
Harness, AI harness, LLM harness, Agent scaffolding

Sources

  • Anthropic: Effective harnesses for long-running agents
  • LangChain: The anatomy of an agent harness
  • OpenAI: Harness engineering

Related terms

  • AgentsAI AgentAn AI system that can autonomously plan and execute actions using tools to achieve a goal.
  • AgentsTool CallingThe broader pattern in which an AI system autonomously selects and requests external software capabilities.
  • Generative AIContext WindowThe maximum amount of information, measured in tokens, that a model can process within one interaction.
  • Generative AISystem PromptThe foundational set of hidden instructions that define an AI model’s persona, boundaries, and behavior.
  • AgentsAI Agent MemoryAn application-level system that stores and retrieves relevant information across steps or sessions for an AI agent.
  • SafetyAI GuardrailsSafety mechanisms and policies put in place to ensure AI systems operate within defined ethical, legal, and operational boundaries.

Read more on Learn

All articles
  • A building under construction wrapped in scaffolding against a blue skyAI Concepts · Nov 3, 2026What Is Harness Engineering? The System Around an AI AgentHarness engineering is designing everything around an AI model so an agent works reliably: instructions, tools, memory, checks and permissions.
  • Illustration of an AI robot working at a deskAI Concepts · Aug 20, 2026What Is an AI Agent? A Beginner-Friendly ExplanationLearn what an AI agent is, how it differs from a standard chatbot like ChatGPT, and how agents use tools to take action in the real world.
  • Illustration of AI connecting to different tools like puzzle piecesAI Concepts · Aug 27, 2026What Is Function Calling? AI Tool Calling ExplainedLearn what function calling (or tool calling) is, how it allows AI models to take real-world actions, and why it is the backbone of modern AI agents.