Claude Code

Claude Code is an agentic coding tool developed by Anthropic that operates inside a developer's terminal, using the Claude family of language models to read and edit files, execute shell commands, search codebases, and carry out multi-step software engineering tasks with limited human supervision. It was announced in February 2025 alongside Claude 3.7 Sonnet and became one of the most prominent examples of the shift from autocomplete-style code assistants toward autonomous AI agents that plan, implement, and verify changes themselves.^[1][2]

History

Claude Code launched as a research preview on 24 February 2025, bundled with the release of Claude 3.7 Sonnet, which Anthropic described as the first hybrid reasoning model capable of both fast responses and extended step-by-step thinking. Claude Code was the first Anthropic product to make systematic use of this agentic capability, operating through a command-line interface where it could be granted permission to modify a repository, run tests, and iterate on failures.^[1]

Following its initial release, Anthropic expanded Claude Code beyond the terminal, adding a web-based interface, desktop applications, and an extension for popular integrated development environments. The tool was also integrated with GitHub Actions, allowing teams to delegate maintenance tasks such as dependency upgrades and pull-request review to automated Claude Code sessions. Anthropic has continued to iterate on the tool's safety and permission model, introducing sandboxed execution for supported runtimes so that the agent's commands can be contained when running untrusted code.^[2][3]

Key Concepts

Claude Code is an agentic coding tool: rather than only suggesting completions, it plans a sequence of actions, invokes tools such as file editing and command execution, observes the results, and adjusts its approach until a task is complete. It is accessed through a terminal with commands such as claude, and can be pointed at a repository to begin working.^[3]

The tool is built on tool use, the ability of Claude models to emit structured calls to external functions. Claude Code ships with a built-in set of tools — file reading and writing, code search, terminal command execution, and web search — and supports custom tools through Anthropic's Model Context Protocol (MCP), allowing developers to connect the agent to their own internal systems and databases.^[3]

Access is tied to Anthropic's subscription tiers: Claude Pro and Max plans include usage of Claude Code, and heavier use is billed through the Anthropic API. Because the agent can execute arbitrary commands on a machine, Anthropic provides permission controls, sandboxing, and audit logs, and documentation warns developers to review the tool's access in production and sensitive environments.^[3][4]

Applications and Impact

Claude Code is used for a broad range of software engineering work: feature implementation, bug fixing, refactoring, test generation, documentation, and repository maintenance. Its multi-step, self-correcting workflow distinguishes it from earlier generation code assistants, and it has been widely benchmarked on agentic coding evaluations derived from SWE-bench, a dataset of real GitHub issues used to measure automated software engineering performance.^[5]

Adoption of Claude Code contributed to the rapid growth of agentic coding as a category in 2025, alongside tools such as Cursor, GitHub Copilot's agent mode, and Google's Gemini Code Assist. For many developers it became the default way to interact with Claude for programming tasks, and Anthropic has described agentic coding as one of its fastest-growing product areas.^[2][6]

>See Also

• Claude
• Agentic AI
• Cursor
• GitHub Copilot
• Devin AI
• Tool Use

References