
Claude has introduced a 'Record a Skill' feature that allows users to teach it tasks by recording their screens and explaining each step. This feature enables the creation of reusable skills that can be used in Claude Chat or Claude Co-work. Users can refine these skills by reviewing the skill.md file, which documents the steps learned. This advancement enhances Claude's utility in automating repetitive tasks, making it a more powerful tool for task management.
Read originalEarlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
Duncan Rogoff · April 23, 2026 · Related
Claude Code Releases · May 12, 2026 · Related
Cole Medin · May 14, 2026 · Related
Matt Wolfe · May 15, 2026 · Background
Skill Leap AI · May 28, 2026 · Background
Claude Code Releases · May 29, 2026 · Related
The AI Daily Brief · June 27, 2026 · Background
Anthropic · July 9, 2026 · Background
Duncan Rogoff · August 13, 2026 · Related
Lev Selector · August 21, 2026 · Background
TechCrunch AI · August 25, 2026 · Background
Cole Medin · September 15, 2026 · Background
Matt Wolfe · October 19, 2026 · Related
Claude Learns Tasks by Watching Users
2 developments
Exin Therapeutics has demonstrated an AI agent physically controlling a robotic arm to inject drugs into a live mouse's brain. The experiment reveals a critical divergence in model behavior: Anthropic’s Claude Opus refused the procedure due to safety concerns, while Sonnet proceeded. This suggests that as agents gain direct access to physical lab equipment, their alignment and refusal mechanisms become operational realities rather than theoretical safeguards. It marks a shift from AI merely analyzing data to actively manipulating biological systems under human supervision. The team initially tested Claude Opus 5.5, but it halted the injection, citing potential harm. Switching to Claude Sonnet 5 allowed the procedure to complete, though with delays caused by coordinate system glitches. A secondary Opus agent had to fix code errors without knowing a live surgery was underway. Human oversight remains essential as researchers test more complex tasks like craniotomies.