Anthropic launched a feature called 'Teach Claude a Skill' on Thursday. It lets AI agents learn by watching your screen. The company says it's a step toward more practical, hands-on AI training. Crypto builders are already using it to write and debug smart contracts.
Watching and learning
The feature works like this: a user shows Claude a task on their screen — navigating a website, filling out a form, writing code. The AI watches, picks up the pattern, and can then repeat the task on its own. Anthropic says the goal is to make AI agents more useful for real-world workflows, not just chat.
Early testers describe it as a mix of screen recording and programming by demonstration. No need to write instructions from scratch. Just show the AI what to do.
Crypto builders get hands-on
Smart contract developers are among the first to experiment. They're using 'Teach Claude a Skill' to train agents to audit code, deploy contracts, and even fix bugs. One developer posted a demo where Claude watched a Solidity contract being written, then caught a reentrancy vulnerability on its own.
The timing isn't accidental. Smart contract exploits have cost the industry billions. A tool that lets AI learn from human experts — and then apply that knowledge — could change how teams handle security.
Anthropic hasn't said when the feature will roll out to all users. For now, it's available to a limited group of testers. But the early results are already circulating on X and GitHub. If the feature works as advertised, it could give crypto developers a new way to automate the tedious parts of building on-chain.
Whether it catches on beyond the crypto crowd is another question. But for a Thursday launch, it's getting attention.
