AI Dev Stack

Sign in Subscribe

anthropic

News zu Anthropic und Claude

Claude Opus 4.8 Now Available in GitHub Copilot

Claude Opus 4.8 Now Available in GitHub Copilot

Claude Opus 4.8 is now generally available in GitHub Copilot for Pro+, Business, and Enterprise users. The model shows clear improvements in code understanding, complex problem-solving, and large-codebase navigation. Rollout is gradual across VS Code, Visual Studio, JetBrains, Xcode, Eclipse, and...

Infrastructure Noise in AI Coding Evals: The 6-Point Leaderboard Gap

Infrastructure Noise in AI Coding Evals: The 6-Point Leaderboard Gap

Anthropic found that infrastructure configuration alone creates a 6-point spread on Terminal-Bench scores — larger than most leaderboard gaps. Resource limits below 3x cause spurious kills; above 3x they help agents solve different problems entirely.

Anthropic's Agent Skills: Composable Expertise for Claude Agents

Anthropic's Agent Skills: Composable Expertise for Claude Agents

Anthropic's Agent Skills let you package domain expertise into composable folders that Claude loads on-demand. Progressive disclosure keeps context lean while giving agents access to specialized knowledge and executable code.

Anthropic's Managed Agents: Decoupling the Brain from the Hands

Anthropic's Managed Agents: Decoupling the Brain from the Hands

Anthropic decoupled Claude from execution environments and session state. The result: 60% faster p50 latency, crash-resistant harnesses, and interfaces designed to outlast any specific implementation as models improve.

Claude Code Gets Sandboxing: 84% Fewer Permission Prompts

Claude Code Gets Sandboxing: 84% Fewer Permission Prompts

Anthropic added OS-level sandboxing to Claude Code, cutting permission prompts by 84% while blocking prompt injection attacks. The sandbox runtime is open source, and Claude Code now runs securely in the cloud.

Code Execution with MCP: Why Your AI Agent Is Wasting Tokens

Code Execution with MCP: Why Your AI Agent Is Wasting Tokens

Loading thousands of MCP tool definitions upfront burns 150,000+ tokens before your agent starts working. Code execution cuts that by 98.7% by loading tools on-demand and processing data locally. Here's how it works.