What are AI coding agents?
AI coding agents are software in which a language model runs in a loop with tools: it reads a codebase, edits files, runs commands and checks results until a goal is met. The category has no single coiner; Simon Willison defines coding agents by one feature, that they can both generate and execute code. It names a tool, not a way of working.
Origin No single coiner is established: Anthropic’s “Building effective agents” (opens in a new tab) of 19 December 2024 gave coding agents a section of their own, Simon Willison (opens in a new tab) defined an agent on 18 September 2025 as an LLM that “runs tools in a loop to achieve a goal”, and his post of 23 February 2026 (opens in a new tab) names coding agents as tools “where the defining feature is that they can both generate and execute code”.
- In Polish
- Agenci kodujący AI
- Closest ladder level
- All levels, free guide
- Also searched as
- coding agents, AI coding agent, agentic coding tools, CLI coding agents
Then $19.99 a month or $99.99 a year. Cancel from your account page. 30-day refund on a first purchase.
Why it matters.
An agent is a tool category, so choosing one tells you where the loop runs, not whether its output is right. Anthropic’s guide to building agents makes the point: “whereas automated testing helps verify functionality, human review remains crucial for ensuring solutions align with broader system requirements.” In practice the agent’s own test runs are a first check, and the decision to merge stays with a person or a gate you defined. Whether anyone reviews the output is a habit of the user, and any of these tools can support either habit.
AI coding agents vs vibe coding
A coding agent is a tool and vibe coding is a way of using one, so the two columns answer different questions.
| Compared on | Vibe coding | AI coding agents |
|---|---|---|
| What it names | A way of working: accepting AI-written code without reading it | A tool category: a model in a loop with tools that reads, edits, runs and checks code |
| Who writes the code | The model | The agent |
| Who checks it | Nobody reads the diff; the human checks behaviour by eye | The agent’s own test runs, then a human: “human review remains crucial” (Anthropic) |
| What stops the loop | “it mostly works” (Karpathy) | The goal is met: “there is a stopping condition” (Willison) |
| Where the name came from | Andrej Karpathy, post on X, 2 Feb 2025 | No coiner; Anthropic’s guide to agents gives coding agents a section, 19 Dec 2024 |
| Where it breaks | Past “throwaway weekend projects”, the scope Karpathy gave it himself | A tool is not a stance: the same agent supports vibe coding or agentic engineering, depending on who checks |
Coding agents beyond Claude Code, Codex and Cursor
This site’s landscape page sorts the others into three groups by where they run, and ranks none of them.
| Group | What the group does | Examples on the page |
|---|---|---|
| Cloud agents | Turn an issue into a pull request | GitHub Copilot, Jules |
| Scriptable terminal agents | Run from a terminal and can be scripted | Gemini CLI, OpenCode, Amp, Factory Droid |
| Editor or desktop environments | The agent lives in an editor or a desktop app | Kiro, Cline, Kilo Code, Junie, Warp |
How each tool does it.
Cursor, Claude Code and Codex each name this their own way. Each note carries the date it was checked.
Cursor
Keeps the agent in the editor, next to Tab completion and inline edit; it also has a CLI and Cloud Agents in isolated VMs.
These cells come from Cursor’s documentation as read on 2026-08-28, because cursor.com could not be reached on 2026-09-26. Re-check a cell before you rely on it.
Cursor docs, 2026-08-28
Claude Code
Starts in the terminal (claude) and runs the same session in the IDE, the desktop app and the web (claude --cloud).
On 2026-09-26 it had subagents, worktrees, skills, hooks (33 events), MCP, a headless mode (claude -p), an SDK and scheduled runs (Routines, in the cloud).
Feature matrix, Claude Code 2.1.283, 2026-09-26
Codex
One agent across the CLI (codex), an IDE extension, the desktop app (codex app) and cloud tasks (codex cloud, experimental).
On 2026-09-26 it had subagents, worktrees, skills, hooks (12 events, which need persisted trust before they run), MCP, a headless mode (codex exec), an SDK and scheduled runs (local automations in the desktop app).
Feature matrix, Codex 0.157.1, 2026-09-26
Questions about AI coding agents.
How are AI coding agents different from vibe coding?
A coding agent is the tool, and vibe coding is one way of using it. Simon Willison keeps vibe coding to “its original definition of coding where you pay no attention to the code at all”, and describes agentic engineering as professionals using coding agents to amplify their expertise. The same agent can serve either, depending on whether a human specifies the work and checks the result. That tool-versus-habit split is our reading.
What is the difference between an AI coding assistant and an AI coding agent?
An assistant suggests code; an agent runs a loop. Faros AI, a vendor, puts it this way: “An assistant suggests the next line or answers questions inside your editor. An agent works more autonomously: it understands a repository, makes multi-file changes, runs tests, and iterates on a task with minimal input.” Anthropic described the same shift in December 2024 as “capabilities evolving from code completion to autonomous problem-solving”.
Is an AI coding agent the same as agentic coding?
No: one is the tool, the other the way of working. A coding agent is software that runs a model in a loop with tools. Agentic coding is working that way, and agentic engineering is the discipline of doing it without lowering the quality bar. Willison uses “agentic engineering” to refer to “building software using coding agents”.
Is a cloud agent a coding agent?
Yes: one that runs the loop in a remote environment instead of your editor or terminal. The feature matrix lists a cloud option for all three tools this site compares: Claude Code web sessions, Codex cloud (experimental) and Cursor Cloud Agents. Vendors rename the category: GitHub’s changelog of 1 April 2026 says “Copilot cloud agent (formerly known as Copilot coding agent)”.
The vocabulary of AI-driven development
Every name below is defined against vibe coding: who writes the code, who checks it, and what stops the loop.
Read the long-form guide: agentic engineering vs vibe coding
Sources.
The primary sources outside this site that this page relies on.
- Building effective agents (opens in a new tab) Erik S. and Barry Zhang, Anthropic
- I think “agent” may finally have a widely enough agreed upon definition to be useful jargon now (opens in a new tab) Simon Willison
- Writing about Agentic Engineering Patterns (opens in a new tab) Simon Willison
- Best AI coding agents for developers in 2026 (real-world reviews) (opens in a new tab) Neely Dunlap, Faros AI
- Research, plan, and code with Copilot cloud agent (opens in a new tab) GitHub, GitHub Changelog
Keep reading.
The guides that go deeper, and the terms and comparisons next to this one.
In the docs
- The full A-Z glossarySubscription
- Vibe coding vs agentic engineering: 21 terms comparedFree
- Tool Comparison OverviewSubscription
- Cursor vs Claude Code vs Codex: feature comparison matrixSubscription
- The coding-agent landscape beyond the big threeSubscription
- Beyond the Big Three: OpenCode, Pi, Copilot CLI, Cline, Goose and Other Coding AgentsSubscription
- The autonomy ladder: which level is your workflow at?Free
Related terms
Read the guides in the same words.
Open every guide with the 7-day free trial. Each term is defined once here, and the guides use it the same way.
Then $19.99 a month or $99.99 a year. Cancel from your account page. 30-day refund on a first purchase.