Anthropic's Claude Cowork Learns New Skills From Screen Recordings and Voice
Anthropic has taught its Claude Cowork desktop app to pick up skills by demonstration. The user records their screen while performing a task and narrates it aloud, and Claude assembles a reusable skill from the recording. Instead of writing out instructions in text, users can now show a task once.
AI-processed from The Decoder; edited by Hamidun News
Anthropic has added a screen-recording learning mode to its Claude Cowork desktop app. As of July 2026, users can record themselves performing a task and layer a voice commentary on top, and Claude will build a reusable skill from the result.
How Claude Cowork learns a skill
Claude Cowork now learns by watching a demonstration rather than reading instructions. The user starts a screen recording inside the app, performs the task manually, and explains aloud what they are doing and why. Claude analyzes the combination of on-screen actions and spoken explanations and packages it as a skill that can later be run again without repeating the steps by hand.
- Developer: Anthropic; product: the Claude Cowork desktop app
- Input: a screen recording made while performing a task
- Addition: voice commentary over the recording
- Output: a reusable skill that Claude applies again
The point is that a demonstration of a single task becomes a procedure that can be used many times: show it once, and the agent reproduces it.
Why this is easier than text instructions
Demonstration removes the main barrier to teaching an agent: having to describe every step precisely in words. Previously, giving Claude a new skill meant writing the procedure out in text: which steps, in what order, what to look at. Now the same result comes from showing the task, by recording the real execution with voice explanations.
Anthropic has been developing skills as a way to store and reuse an agent's repeatable procedures. Screen recording, reported by The Decoder, extends that format to tasks that are easier to show than to describe: working with interfaces, multi-step operations across apps, and routine work where the order of clicks matters.
"Claude
Cowork now lets you record your screen as you perform a task, add voice commentary, and turn all of it into a reusable skill," is how The Decoder describes the new feature.
What this means
Training an AI agent is shifting from writing instructions to demonstrating. Showing a task becomes enough for the agent to reproduce it. For desktop agents, this lowers the barrier to entry: anyone can build a skill, not just people who can formulate procedures in text. Anthropic is building this style of teaching directly into Claude Cowork, moving the agent closer to a "do it the way I showed you" workflow.
- Oct 28, 2026A 2015 CNN Beat GPT-6 Astra and Claude Fable 5.1 at Skin Cancer Diagnosis
- Oct 28, 2026Keystone removes the model name from code: open-source LLM gateway adds 0.64 ms
- Oct 28, 2026Wired Tested 6 AI Agents in a Day: Dots Leads, Gemini Spark Fails at Lunch
- Oct 28, 2026Anthropic files for IPO: $16.3 billion in revenue in six months, ticker ANT
Key Questions Answered
How Claude Cowork learns a skill?
Claude Cowork now learns by watching a demonstration rather than reading instructions. The user starts a screen recording inside the app, performs the task manually, and explains aloud what they are doing and why. Claude analyzes the combination of on-screen actions and spoken explanations and packages it as a skill that can later be run again without repeating the steps by hand. Developer: Anthropic; product: the Claude Cowork desktop app Input: a screen recording made while performing a task Addition: voice commentary over the recording Output: a reusable skill that Claud
Why this is easier than text instructions?
Demonstration removes the main barrier to teaching an agent: having to describe every step precisely in words. Previously, giving Claude a new skill meant writing the procedure out in text: which steps, in what order, what to look at. Now the same result comes from showing the task, by recording the real execution with voice explanations. Anthropic has been developing skills as a way to store and reuse an agent's repeatable procedures. Screen recording, reported by The Decoder, extends that format to tasks that are easier to show than to describe: working with interfaces, m
What this means?
Training an AI agent is shifting from writing instructions to demonstrating. Showing a task becomes enough for the agent to reproduce it. For desktop agents, this lowers the barrier to entry: anyone can build a skill, not just people who can formulate procedures in text. Anthropic is building this style of teaching directly into Claude Cowork, moving the agent closer to a "do it the way I showed you" workflow.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.