Your daily AI digest for developers — Friday, April 03 2026
Cursor has launched a new AI agent experience that aims to compete directly with OpenAI's Codex and Anthropic's Claude Code. This new experience focuses on enhancing the capabilities of AI agents in coding tasks.
This tutorial provides a step-by-step guide to building AgentScope workflows using ReAct agents and custom tools, focusing on structured output and concurrent pipelines.
A leaked version of Claude Code source code was found to contain malware, posing significant security risks to developers who downloaded it.
The article discusses how Google's Gemma 4, NVIDIA, and OpenClaw are optimizing local agentic AI, reducing costs associated with running AI models.
GitHub has launched an AI-powered workflow to manage accessibility feedback, using GitHub Actions, Copilot, and Models APIs to centralize and automate issue management.
Microsoft has released three new foundational AI models focused on speech recognition, speech synthesis, and image generation, enhancing its competitive edge in AI development.
Microsoft's Terms of Use for Copilot acknowledge that the tool is intended for entertainment and may produce incorrect results, highlighting the need for caution when using AI-generated code.
Google DeepMind's Gemma 4 models are now available with Apache 2.0 licensing, offering powerful open models for reasoning and coding tasks.
Trinity Large Thinking is a new open-source reasoning model designed for long-horizon agents and tool use, supporting complex multi-step reasoning tasks.
New Rowhammer attacks can compromise machines with Nvidia GPUs, highlighting a significant security vulnerability in AI and computing environments.