AI Coding Agents
Tools that plan, write, edit and run code on a developer's behalf.
AI Coding Agents compared: browse and filter
Filters and search update this list on the page and do not create new URLs.
-
Claude Code
Anthropic · Agentic coding tool (terminal-first) Agentic tool
Anthropic's agentic coding tool for the terminal, IDEs, a desktop app and the web, built around permissions, subagents and MCP.
- Vendor-sourced
Key capabilities- Terminal CLI
- IDE extensions
- Desktop app and web
- Permissions
PricingIncluded with paid Claude plans, or usage-based via API billing.Structure only; figures not recorded -
Cursor
Anysphere · AI code editor with agent mode Agentic tool
An AI code editor with an agent, Tab completion, cloud agents and code review, offering models from several vendors.
- Free plan
- Free trial
- Vendor-sourced
Key capabilities- Agent
- Tab completion
- Max Mode
- Cloud agents
PricingFree Hobby plan; Pro $20 per month.Prices from a vendor-published source
No listings match these filters.
Listings are ordered alphabetically by default. They are not ranked, and there are no user ratings or paid placements.
What is AI Coding Agents?
An AI coding agent is software that uses large language models together with tools, such as file editing and shell commands, to carry out programming tasks. Most current products are human-directed: the developer reviews or approves changes, which is why we describe them as agentic tools rather than fully autonomous agents.
AI coding agents go beyond autocomplete. Given a task, they can read a repository, plan changes across several files, run commands such as tests, and propose or apply edits for a developer to review.
Products in this category differ in where they run (a purpose-built editor, a terminal, an IDE extension or a cloud environment), which models they use, how much they act without asking, and how usage is billed. Those differences matter more than headline capability claims.
We classify each listing as an agentic tool, an assistant or an autonomous agent only where vendor documentation supports it, and we label anything not documented as unknown.
The complete guide to AI Coding Agents
AI Coding Agents compared on key criteria.
Summaries come from each profile; open a profile for sources and caveats.
| Product | Interface | Model choice | Free tier | Pricing model | Human oversight |
|---|---|---|---|---|---|
| Claude Code | Terminal, VS Code, JetBrains, desktop app, web | Claude models (Anthropic) | Not documented in sources reviewed | Included in paid Claude plans, or usage-based API billing | Permissions, sandboxing and checkpoints documented |
| Cursor | AI code editor (desktop) with cloud agents | OpenAI, Anthropic, Google and Cursor models | Yes (Hobby) | Monthly tiers; usage at model API rates | Not documented in sources reviewed |
Which AI Coding Agents option fits which situation.
- Claude CodeMay suit: Terminal-first developers.
- CursorMay suit: Developers who prefer an editor-centred workflow.
Based on each profile's documented strengths and limits. These are fit notes, not rankings: test options against your own requirements.
Who uses AI Coding Agents and what problems it solves.
Individual developers
Use agents to scaffold features, refactor code, write tests and explore unfamiliar repositories faster.
Engineering teams
Share rules, commands and review workflows so that agent output follows team conventions.
Platform and security teams
Evaluate data handling, permission controls, SSO and central configuration before wide rollout.
Teams with large or legacy codebases
Care about context management, repository navigation and how well an agent handles changes that span many files.
Features and criteria to evaluate in AI Coding Agents.
Interface and workflow fit
Decide whether you want an AI-native editor, a terminal tool, an IDE extension or cloud-run agents, and whether your team can switch tools.
Model choice
Check whether the product offers several model vendors or one family, and how model access changes by plan.
Permissions and human oversight
Look for documented controls over file access, command execution and approvals, plus checkpoints or review steps.
Context handling
Review how the tool provides project instructions, memory and large-context options for big repositories.
Pricing and usage limits
Understand whether usage is flat, credit-based or metered at model API prices, and what happens when included usage runs out.
Data handling and administration
Read the vendor's privacy and training policies, and check for org-wide controls and SSO.
How to choose AI Coding Agents: a step-by-step approach.
Pick the interface your team will actually use
Decide between an AI-native editor, a terminal tool, an IDE extension or cloud-run agents, and whether developers can switch tools.
Define the tasks
Separate autocomplete, multi-file edits, test fixing, code review and migrations, because tools differ in how well they cover each.
Review model and data policies
Check which models are offered, whether your code can be used for training, how long data is retained, and which enterprise controls exist.
Check oversight and permissions
Look for documented approvals, permission rules, sandboxing and checkpoints, and decide what the agent may run without asking.
Estimate cost under real usage
Understand included usage, metered or on-demand charges and rate limits, then measure consumption on real tasks.
Run a bake-off on your own repository
Give each candidate the same tasks on a branch, and compare correctness, review effort, cost and developer preference.
Common mistakes when choosing AI Coding Agents.
Judging on demos
Demos use clean projects. Evaluate on your own codebase, conventions and test suite.
Ignoring review effort
Faster code generation is not a gain if reviewing and fixing the output takes longer than writing it.
Skipping the data policy
Check how your code and prompts are handled before connecting a private repository.
Leaving permissions wide open
Decide in advance which commands and files an agent may touch without approval.
Assuming one tool fits every workflow
Editor-centred and terminal-first tools suit different habits, and teams often need to test both.
AI Coding Agents key terms explained.
- Agentic coding
- Using an AI system that plans and carries out multi-step programming tasks with tools, rather than only suggesting code.
- Context window
- The amount of text a model can consider at once. Large repositories usually exceed it, so tools must select what to include.
- MCP (Model Context Protocol)
- An open protocol for connecting AI tools to external data sources and tools.
- Subagent
- A helper agent with its own context and tools that handles a sub-task and reports back to the main session.
- Sandboxing
- Restricting what an agent's commands can access on a machine so mistakes or unwanted actions have limited impact.
- Human-in-the-loop
- A workflow in which a person reviews or approves the agent's actions before they take effect.
AI Coding Agents questions and answers
Is an AI coding agent the same as an AI code assistant?
Not exactly. An assistant mainly suggests code or answers questions. An agent can also take actions, such as editing several files or running commands, usually under user approval. Products often combine both modes.
Are AI coding agents fully autonomous?
The tools listed here are directed by a developer who reviews or approves changes. We only describe a product as autonomous when its vendor documentation says so.
How should I test an agent on a large codebase?
Run the same real task, such as a cross-module refactor with tests, in each candidate on a branch of your own repository, and compare correctness, review effort, cost and how well the tool handles project context.
What should security teams check first?
Start with the vendor's data-retention and training policy, which commands and files the tool can access, whether permissions can be enforced centrally, and the availability of SSO and audit controls.
Narrow your AI Coding Agents shortlist.
Open a profile for sources and pricing, compare two products side by side, or read how others approach the decision.
Logos provided by Logo.dev. Product names and logos are trademarks of their respective owners and appear for identification only; they do not imply endorsement.