AI agent
Agents and tools · LLM agent, agentic AI
An AI agent (LLM agent, agentic AI) is a program where an LLM works inside a harness: it picks its own steps, reads files, runs commands, calls tools and checks the result right inside your environment. Example: Claude Code gets the task “fix the failing test”, reads the code, makes an edit, runs the tests and repeats until the test passes.
Hover a zone or press ▶ Step
What an AI agent is made of
| LLM | Harness | AI agent | |
|---|---|---|---|
| LLM | ✓ | ✗ | ✓ |
| Agent loop | ✗ | ✓ | ✓ |
| Tool and tool calling | ✗ | ✓ | ✓ |
| MCP | ✗ | ✓ | ✓ |
| Context | ✗ | ✓ | ✓ |
| Agent memory | ✗ | ✓ | ✓ |
| CLAUDE.md, Project bible, Skill | ✗ | ✓ | ✓ |
| Permissions | ✗ | ✓ | ✓ |
| Sandbox | ✗ | ✓ | ✓ |
| Hooks | ✗ | ✓ | ✓ |
| Static checks | ✗ | ✓ | ✓ |
| Human in the loop | ✗ | ✓ | ✓ |
| Orchestration | ✗ | ✓ | ✓ |
| Logs and tracing | ✗ | ✓ | ✓ |
LLM agent architecture: LLM + harness (loop, tools, context, rules, checks).✓has✗none
AI agent examples · 5
- Claude Code · A coding agent from Anthropic: it reads the project, edits files, runs commands and tests, and works with git.
- Codex · A coding agent app from OpenAI on GPT models: it reads the project, edits files, runs commands and tests, and prepares pull requests.
- Cursor · A code editor app from Anysphere, built on VS Code, with a built-in AI agent: you work in the editor, the agent sits next to you, and you pick a model from different vendors or use its own Composer model.
- Hermes Agent · A free, open-source agent from Nous Research that you install on your own server or computer and connect to any model, through a vendor API or a local one.
- Yandex · A Russian company that builds the Alice AI language models, and earlier the YandexGPT line.
Source: Anthropic: Building effective agents
Read more: AI agent = LLM + harness: LLM vs harness vs agent
Where it appears · 22
- AI agent = LLM + harness · Harness
- MCP server and AI agent for 1C · Harness
- How many AI agents you need · Harness
- Will AI agents take your job · Harness
- Analytical reports with an AI agent · Harness
More 17
- Claude in Chrome · Harness
- The agent’s data plane and control plane · Harness
- Incident investigation with an AI agent · Harness
- MCP scopes in Claude Code · Harness
- Requirements and meeting notes with AI · Harness
- QA work after support with an AI agent · Harness
- MCP for an AI agent · Harness
- Claude Code, Codex, Cursor and Hermes Agent · Harness
- How to run and analyze interviews about AI use · Research
- ChatGPT Enterprise usage report for a company · Research
- Harness engineering in practice · Harness
- Approaches to working with a harness · Prompts and review
- A reconcile loop for specs · Specs and context
- Specs by aggregate · Specs and context
- AGENTS.md and CLAUDE.md, specs, rules and skills · Specs and context
- Who to interview before buying AI tools · Research
- Prompts for work tasks · Prompts and review