Claude Code is an agentic coding tool developed by Anthropic and built around its Claude models. Rather than only answering programming questions in a chat window, it works inside a software project: it inspects source code, modifies files, runs commands and checks its own results. Anthropic's documentation describes it as available in a terminal, an IDE, a desktop app and a browser. It belongs to the category of AI agents, which combine a large language model with the ability to take actions through software tools.
History
A third-party history of the product traces its origin to a side project started by one engineer in September 2024, which entered a limited research preview in February 2025 and was reported to pass a $1 billion annualized run rate by November 2025. These figures come from secondary reporting and have not been independently verified by Anthropic.
In March 2026 the source code of the command-line application leaked. The leak revealed multiple upcoming features and models.
On 6 May 2026 Anthropic presented updates at its developer event "Code with Claude" in San Francisco, where livestream sessions covered work across Claude Code, the Claude Developer Platform, and partner deployments. At the same event the company raised usage allowances, announcing increased rate limits for developers on Claude Code and the API, including a doubled five-hour limit for Pro, Max and Enterprise customers.
In October 2026 Linux creator Linus Torvalds reportedly disclosed that the tool had helped draft kernel patches and test programs for page-cache read performance, working under his direction. Coverage of the disclosure noted that it also openly documented technical limits. The same month, Anthropic shipped version 2.1.296, adding a workflow-level control that lets teams assign one chosen model to every subagent in a pipeline.
How it works
Model and harness
Commentary often separates the product from the models it uses. The tool is described as the harness, with Opus and Sonnet as the models that run inside it; users select the model with a /model command, while the same agentic loop continues underneath.
Tools and the agent loop
Tools distinguish Claude Code from a conventional chatbot. Without tools, the model can only respond with text, whereas with tools it can read code, edit files, run commands, search the web and interact with external services. Each tool result is returned to the loop and informs the model's next decision. The built-in tools fall into five categories, each representing a different kind of agency.
Codebase navigation and verification
It uses agentic search to infer project structure and dependencies, so users need not select context files manually. Before changing code, the agent searches the repository, reasons through the change, applies it across files, and then runs tests and linters to check its work.
Because it can use a shell, it can commit changes to git, run a build and read the resulting error output. When it needs direction, it can surface decisions the user must make before it continues.
Autonomy
Training material from AI education provider DeepLearning.AI states that the tool can plan, execute and improve code with minimal human input for more than a few minutes.
Interfaces and deployment
The product has grown from a command-line origin into several surfaces. It began as a command-line interface offering the most control, then gained an IDE interface for following code changes more easily, and most recently a desktop application with a full graphical interface. The IDE and desktop versions are built on the Claude Agent SDK, which external developers can also use.
The command-line interface remains the core form. Third-party analysis notes that, unlike IDE-integrated assistants, the tool is designed as a composable Unix utility, so data can be piped into it, several instances can run in parallel, and it can be placed in automated pipelines. Integration with version control hosting is also central: it works with GitHub, GitLab and command-line tools to read issues, write code, run tests and open pull requests. Sessions can also be steered from the terminal, an IDE, Slack or the web.
In March 2026 computer use was added for Pro and Max users, letting Claude open files, run development tools and navigate the screen. A separate product, Claude Cowork, is similar to Claude Code but has a graphical user interface aimed at non-technical users.
Extensibility and configuration
Project-level instructions are stored in a file named CLAUDE.md, which contains guidelines about the codebase that Claude Code can remember across sessions. Integration with external services uses the Model Context Protocol; MCP support connects the tool to external tools, databases and services. The tool can also manage several parallel workstreams, since projects group related coding sessions so that a user can run and supervise several agents at once. Lifecycle hooks have been expanded as well, with more hook events and better streaming behavior added in 2026.
Users can choose the underlying model; models can be switched during a session with /model or set at startup with a command-line flag.
Access and pricing
On first run, users sign in with a Claude subscription (Pro or Max) or an Anthropic API key. Usage limits apply, and prices and plans may change at Anthropic's discretion.
Adoption
Anthropic has cited large enterprise users. At the May 2026 event it named companies including Shopify and Mercado Libre, and said Mercado Libre aimed for "90% autonomous coding by Q3 this year." These are company claims and have not been independently assessed.
Competing products have appeared in the same category. OpenAI and Anthropic released competing command-line coding agent updates within 48 hours of each other in October 2026.
Limitations and controversies
Public understanding is complicated by naming. Some users conflate the tool with the chat product: Claude Code is Anthropic's agentic coding tool, and it is not the Claude people chat with.
Because the tool can run commands and modify files, its actions depend on permissions. It operates under permissions that the user controls. Granting an agent shell access raises security concerns, and the leak of its source code in 2026 drew attention to the software's internal design. Agent-generated changes also require human review, a point that users and reviewers commonly emphasise.
References
- Overview - Claude Code Docscode.claude.com
- What Is Claude Code? Full History, Timeline & How It Workstaskade.com
- Claude (language model)en.wikipedia.org
- Anthropic's Code with Claude Announces Managed Agents, Proactive Workflows, Capability Curve - InfoQinfoq.com
- Live blog: Code w/ Claude 2026simonwillison.net
- Linus Torvalds Reveals Claude Code Wrote Linux Kernel Patches and Test Programscybersecuritynews.com
- Codex CLI 0.162 vs Claude Code 2.1.296 [2026] - Tech Insidertech-insider.org
- What Is Claude Code? Anthropic's Agentic Coding Tool, Explainedgeotoolbox.ai
- How Claude Code works - Claude Code Docscode.claude.com
- Claude Code by Anthropicclaude.com
- Claude Code — Anthropic's official CLI coding agentopenagents.org
- Claude Code: A Highly Agentic Coding Assistant - DeepLearning.AIdeeplearning.ai
- Inside Claude Code: A Deep Dive into Anthropic’s Agentic CLI Assistantmedium.com
- Every Claude Code Update From March 2026, Explainedbuilder.io
- GitHub - anthropics/claude-code: Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands. · GitHubgithub.com