Anthropic turns to ‘skills’ to make Claude more useful at work

AI agents spent years as a concept and then as an experiment. Now, AI companies are devoting even more time and resources than before to make their agents actually useful for end users, whether they’re consumers or professionals.

Anthropic on Thursday announced its next step toward that goal: Skills for Claude. The tool is made up of “folders that include instructions, scripts, and resources that Claude can load when needed to make it smarter at specific work tasks — from working with Excel [to] following your organization’s brand guidelines,” per a release. People can also build their own Skills for Claude relative to their specific jobs and use them across Claude.ai, Claude Code, Anthropic’s API, and the Claude Agent SDK. Box, Rakuten, and Canva have already used the tool, according to the release.

Essentially, the feature is designed to improve Claude’s AI agent capabilities for your work specifically, so you don’t have to spend as much time writing the perfect prompt or referring to past context every time you’re trying to accomplish a task. It’s available to Pro, Max, Team, and Enterprise users.

Brad Abrams, a product lead at Anthropic, told The Verge that “the thing that’s interesting to me about Skills is basically about agents.” He said that Skills as a feature essentially provides organizations building agents a way to teach Claude to do a good job “in their specific context.” He emphasized that it’s not about meeting arbitrary benchmarks — it’s about being able to do the task you need at your own company.

Using an Anthropic layer on top of Claude’s PowerPoint Skill, “I had Claude create me a presentation about how Haiku 4.5 is doing in the market,” Abrams said, adding that Claude created “well-formatted slides that are easy to digest.”

OpenAI, Anthropic, Google, Microsoft, and others have been working toward the goal of actually useful AI agents for years, with executives regularly bringing up agentic AI on earnings calls and redirecting internal resources toward building the tools. To date, though, progress has been largely incremental, with companies fighting to release new feature updates or iterations of agents. (Think: Anthropic’s Computer Use — or OpenAI’s Operator, then Deep Research, and then ChatGPT Agent, which essentially combined the two.)

Anthropic’s news also follows an OpenAI announcement in the same realm earlier this month at the company’s annual DevDay event.

At the event, OpenAI unveiled AgentKit, a group of tools executives said were “designed to help you take agents from prototype to production” and targeted both big companies and individual developers. The example use case OpenAI demonstrated was Albertsons, which runs more than 2,000 US grocery stores, using a custom agent with custom data to create a plan to improve ice cream sales if they were down more than 30 percent. Box, Canva, Evernote, and Ramp were also mentioned as having tried the tool. OpenAI also announced a consumer-facing tool that allows people to work with apps inside ChatGPT, like Zillow and Uber Eats.