According to Claude on X (@claudeai), Claude Code now supports computer use, enabling the model to open apps, click through UI, and test what it builds directly from the CLI, available in research preview for Pro and Max plans. As reported by Claude on X, this expands Claude’s agentic capabilities from code generation to end-to-end software execution and validation, which can streamline QA workflows, smoke tests, and regression checks for engineering teams. According to the announcement video cited by Claude on X, the feature suggests tighter integration with local or virtualized app environments, creating opportunities for DevOps teams to automate UI testing and post-deploy verification without bespoke test harnesses. As reported by Claude on X, access is limited to Pro and Max subscribers in research preview, indicating early feature gating and potential enterprise pilots focused on developer productivity and autonomous software testing.

According to God of Prompt on Twitter, Andrej Karpathy's viral AI coding rant was transformed into a system prompt designed to optimize agentic coding workflows, especially for Claude. The prompt focuses on reducing common LLM coding mistakes such as unchecked assumptions, overcomplicated code, and lack of clarification, by enforcing a structured, senior-engineer mindset. As reported by Karpathy, this approach has led to a dramatic shift in software engineering, with engineers now predominantly coding through agentic LLMs like Claude and Codex, moving from manual coding to high-level orchestration. The underlying business opportunity lies in leveraging these new AI-driven workflows to accelerate development, enhance code reliability, and increase productivity, while also preparing organizations for a rapid industry-wide transformation in 2026.