On May 22, 2025, Anthropic made Claude Code, its terminal-based agentic coding tool, generally available after three months as a research preview. The announcement came alongside the launch of Claude Opus 4 and Claude Sonnet 4, and it added beta extensions for VS Code and JetBrains IDEs, background tasks through GitHub Actions and an SDK for building custom agents on the same core. General availability marked the point where agentic coding in the terminal moved from experiment to a supported product that teams could standardize on.
Key Facts#
- GA date: May 22, 2025, announced in Anthropic's Claude 4 launch post. Version 1.0.0 of the
@anthropic-ai/claude-codenpm package was published the same day. - Origin: Claude Code debuted on February 24, 2025 as "a command line tool for agentic coding" in a limited research preview, released together with Claude 3.7 Sonnet.
- IDE integration: New beta extensions for VS Code and JetBrains run Claude Code inside the editor and show proposed edits inline in the files.
- Automation: Claude Code can run background tasks through GitHub Actions, and a beta GitHub integration lets developers tag Claude Code on pull requests.
- Extensibility: An extensible Claude Code SDK exposes the agent's core so developers can build their own agents.
- Models: Claude Opus 4 was priced at $15/$75 and Claude Sonnet 4 at $3/$15 per million input/output tokens.
What Happened#
Anthropic's Claude 4 post was mostly about the new models, but it contained a short, consequential line for developers: Claude Code was now generally available. The release addressed the main limitation of the preview, which lived only in the terminal. With the new IDE extensions, a developer could start Claude Code from VS Code or a JetBrains IDE and review the agent's suggested changes inline, next to the code they affected, instead of reading diffs in a console.
The second change moved the agent off the developer's machine. Through GitHub Actions, Claude Code could run background tasks in a repository, and with the beta GitHub integration a developer could tag Claude Code directly on a pull request. That turned the tool from a pair programmer into something closer to a teammate that picks up work asynchronously.
The third change was the Claude Code SDK. Anthropic described it as a way to build custom agents using the same core technology that powers Claude Code, which invited teams to reuse the agent loop for internal automation rather than re-implementing tool use, file editing and context handling themselves.
The same announcement added four capabilities to the Anthropic API: a code execution tool, an MCP connector, a Files API and prompt caching for up to one hour. Anthropic also published benchmark claims for the new models, including 72.5% on SWE-bench for Opus 4 and 72.7% for Sonnet 4, and 43.2% on Terminal-bench for Opus 4. These are vendor-reported results. Our Claude Opus 4 and Sonnet 4 news story covers the models in detail.
Background#
When Anthropic introduced Claude Code in February 2025, it described a tool that could search and read code, edit files, write and run tests, commit and push to GitHub and use command-line tools, while keeping the developer in the loop at every step. It was released as a limited research preview, and npm records show the first public package on February 24, 2025.
The design choice that set Claude Code apart was its home in the terminal. Rather than wrapping an editor, it worked with the same tools a developer already uses: the shell, the build system, the test runner and git. That made it language- and editor-agnostic, and it gave the agent a natural way to check its own work by running builds and tests.
By spring 2025, agentic coding tools were multiplying, and the question for engineering leaders was shifting from "can an agent write code?" to "can we operate one safely at team scale?" General availability, IDE integration and CI automation were Anthropic's answer to that second question.
Why It Matters for Developers#
For .NET teams, Claude Code's terminal-first design fits well with how the platform already works. The dotnet CLI gives an agent fast, deterministic feedback: dotnet build surfaces compiler errors and analyzer warnings, and dotnet test tells it whether a change broke behavior. The more of your quality bar is encoded in those commands, the better any coding agent performs.
- Make the build strict. Treat warnings as errors, enable nullable reference types and run analyzers in CI. An agent iterates until the build is green, so a strict build pulls its output toward your standards. Our code quality guide shows how to configure this.
- Invest in fast tests. A reliable unit test suite is the agent's best specification. See the unit testing guide for patterns that keep tests fast and deterministic.
- Treat CI agents like any other automation. Background tasks in GitHub Actions need scoped tokens, protected branches and required reviews. The CI/CD with GitHub Actions guide covers the controls to put in place first.
- Review agent output as you would a colleague's. Inline diffs in the IDE make review easier, but they do not replace it. Small, focused tasks produce reviewable changes.
The broader lessons, including when to delegate work to an agent and how to structure prompts and repositories for it, are covered in our guide to AI-assisted .NET development.
What's Next#
At GA, Anthropic positioned the SDK and GitHub integration as the path to running Claude Code beyond an interactive session. That direction continued. On September 29, 2025, Anthropic renamed the Claude Code SDK to the Claude Agent SDK, saying its teams were using it for work well beyond coding, such as deep research, video creation and note-taking. Anthropic describes the SDK's loop as gathering context, taking action, verifying the work and repeating, with support for subagents, automatic context compaction and Model Context Protocol integrations.
The terminal product also kept evolving. Today the project's README lists native installers for macOS, Linux and Windows, including WinGet, and describes the npm installation route as deprecated. It also notes that the tool works in the terminal, in the IDE or through @claude mentions on GitHub. The npm package itself has moved past version 2.1.
Open questions that GA did not settle include how to measure the real productivity impact of agents on large, mature codebases, and how to govern agents that can run commands with a developer's credentials. Those remain active topics for engineering teams in 2026.