Claude Cowork now learns tasks from your screen recordings and voice-over

Anthropic just shipped a feature that nudges Claude closer to a real digital coworker: the ability to learn repeatable tasks by watching you do them once. On Tuesday, the company added a “Record a skill” option to the Claude desktop app’s Cowork workspace. You hit record, do a task on screen while explaining what you’re doing out loud, and stop. Claude converts that screen capture and voice narration into a reusable skill that lives in your library. Next time the same chore comes up, you invoke the skill instead of clicking through every step again.

This isn’t entirely new territory. OpenAI’s Codex tool has offered similar workflow recording and replay capabilities. But Anthropic’s implementation puts the feature directly inside its main desktop interface and makes voice narration the primary teaching mechanism — a design choice that feels less like programming macros and more like training a new hire. The feature sits behind the ”+” menu in Cowork and is currently available to Pro, Max, and Team plan subscribers. No word yet on whether Enterprise users get access or if there’s a limit to how many skills you can save.

The real question is reliability. Screen recording-based automation is brittle by nature — change one UI element or encounter a dialog box that wasn’t there during recording, and the whole sequence can break. Anthropic hasn’t published any benchmarks on skill execution robustness, and the company’s announcement didn’t address how Claude handles unexpected states during playback. That’s the difference between a genuinely useful tool and a demo that impresses until you actually depend on it.

Still, there’s a throughline here worth tracking. Anthropic has been steadily building toward an agent that doesn’t just generate text but operates software. Last month’s computer use feature let Claude control a mouse and keyboard; this new skill recording closes the loop by letting users teach it their specific workflows. If the execution proves solid, the value proposition shifts from “Claude can help you think” to “Claude can do the boring parts of your job while you focus on everything else.” For knowledge workers drowning in repetitive digital tasks, that’s the promise that actually moves the needle. Whether Anthropic has solved the reliability gap is what will determine if this feature sticks or fades into the graveyard of cool-but-flaky AI demos.