Claude Code: the start of agentic coding

2025-02-24 · Publisher : Anthropic · Domain : Artificial Intelligence

Alongside Claude 3.7 Sonnet, Anthropic opens a research preview of Claude Code, an agent that works in the terminal: it reads code, edits files, runs tests and pushes to GitHub.

On February 24, 2025, Anthropic releases Claude 3.7 Sonnet, its first hybrid reasoning model, together with a tool in limited research preview: Claude Code. Where coding assistants completed lines in the editor, Claude Code runs in the terminal and takes on whole tasks. ### What happened - **Capabilities**: according to the announcement, Claude Code can "search and read code, edit files, write and run tests, commit and push code to GitHub, and use command line tools", keeping the developer in the loop at every step. - **Claimed gains**: in early testing, the tool completed in a single pass tasks that would normally take more than 45 minutes of manual work (vendor figure). - **General availability**: on May 22, 2025, with Claude Opus 4 and Claude Sonnet 4, Claude Code becomes generally available. Anthropic adds beta extensions for VS Code and JetBrains, an SDK to build your own agents and background tasks through GitHub Actions. ### Why it matters Claude Code sets up a new way of working: the agent explores the repository, plans, runs commands and checks its work with tests, while the human sets the goal and reviews. This "agent in the terminal" format was then adopted by most players (Codex CLI, Gemini CLI) and fed shared conventions such as project instruction files, MCP servers and Agent Skills. ### Caveats The productivity gains cited are the vendor's. The agent's autonomy calls for guardrails: command permissions, review of changes, protection of secrets. ### Key takeaways February 2025 marks the move from code assistance to delegating development tasks to an agent, with the terminal as the entry point.

Tags : IA, Agents, Code, Anthropic, Claude