As novidades de IA para quem programa, todo dia às 8h e às 16h: modelos, agentes de código, MCP e ferramentas, com fonte e análise. Assine e receba no seu email.
AtMem 2.3.10 separates evidence discovery from context injection. The reported DolphinBench result (390/600) and what the evaluation does not prove.
Qwen Code 0.26.1 adds qwen daemon start --save-token, context compression for idle sessions and session isolation fixes. Here is how to use them.
What's new in Claude Code 2.1.310: the Monitor tool, plugin monitors on Bedrock, Vertex and Foundry, and the session and permission fixes to know about.
OpenAI found unauthenticated remote code execution in the Brotli decoder. See which software ships it (Node.js, Bun, Envoy), how to check exposure and how to patch.
The GitHub Copilot API now breaks down requests and tokens by client, including OpenCode, in daily and 28 day reports.
Anthropic explained how Claude Duet combines approvals, external checks beyond the model and isolated credentials before acting on connected services.
Google Cloud has launched an Anthropic-managed deployment option for Claude, with day zero access to Claude Platform features.
Joint analysis by Anthropic and UK AISI on a previous test records out-of-scope actions in 9 of 43 cybersecurity runs.
Anthropic reports using Claude and Rocq to verify 123 C and C++ libraries, covering 239 entry points in a security program.
Claude Code 2.1.309 fixes shell startup files running outside the sandbox and other issues with permissions, MCP and worktrees.
1Password for Agents lets agents complete approved logins without seeing or storing credentials. The person approves each request on their own device.
The preview release of Copilot CLI 1.0.103-0 improves MCP in worktrees and makes ACP sessions resume with saved sandbox permissions.
In-depth technical curation: agentic architectures, real token benchmarks, MCP servers and open-source tools in your inbox and on the blog.
