Anthropic releases Agent Skills

Edition 129 | October 20, 2025

Imagine being just seconds away from fixing a bug you’ve been chasing all day, then suddenly you have to hear about how GEICO can save you 15% or more on car insurance.

Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!

In today’s issue…

• Claude gets Agent Skills

• Visa and Mastercard go all-in on agent payments

• GPT-5 gets super-sized, Claude 4.5 gets super-cheap

• The biggest names in agents assemble at Open Agent Summit

• A no-code, drag-and-drop builder for web scraping agents

…and more

🔍 SPOTLIGHT

There’s endless debate over which company’s models are the smartest, but at least on paper, Anthropic’s are now the most “skilled”.

Last Thursday, Anthropic launched a new feature for its Claude AI assistant called Agent Skills (or just “Skills”), which allows users to create pre-made guides for carrying out almost any task. These playbooks are stored in a SKILL.md file, and are intelligently loaded into memory by Claude when the skill is needed. In one of the examples given by Anthropic, a games company provides Claude with skills for applying their brand guidelines and creating presentations, which it then decides to utilize when asked to create a pitch deck for the company. In addition to the SKILL.md text file, skills can include other relevant file types like images and code.

Getting meta, one of the skills provided is “skill-creator”, which gives Claude instructions on how to create other skills, allowing users to interactively generate new playbooks that Claude can utilize later. Anthropic also released a longer guide for skill building targeted towards developers.

Skills are now available to all Claude subscribers at the Pro ($20/month) tier and up. Any time one of the major LLM providers launches a new feature, the others almost always follow suit quickly, so expect to see something similar to ChatGPT and Gemini before long.

To me, “skills” feels like a misnomer—a skill is an ability you’ve internalized to the point that you don’t need to consult a reference; it’s just second nature. LLMs already have many skills, like writing code or poetry, that they don’t require additional instructions to perform. Claude’s “skills” are more like guides, playbooks, or instruction manuals. Maybe those names didn’t sound anthropomorphic enough.

So why is Anthropic launching this feature now?

One thing that’s become apparent over the last year or so of the AI race is that the limiting factors to AI agent performance are increasingly shifting from the raw intelligence of LLMs—which has increased by leaps and bounds—to integration and context.

Integration is the ability of agents to access different pieces of software—an agent that reads emails from your clients and saves them to your CRM is useless if it can’t access your email or CRM, obviously. Humans can interface with any piece of software through desktops and web browsers, but agents need custom connectors for each one (though that’s changing). The arms race for integrations is one of the hottest areas of competition among the major AI providers right now.

The new Skills feature, on the other hand, addresses context: giving agents the information they need to complete their task. Sure, LLMs can write computer code, but they might not use the right libraries for your company’s code base if they haven’t been explicitly told which to use and which to avoid. This is where guides…sorry, “skills”, come in. Like a newly-hired software engineer who knows how to code, but nothing about the systems of the company they’re joining, Claude can be given an instruction manual that provides them with the context they need.

Whatever word you want to use for that, it’s an important step forward.

Always keep learning and building!

—Michael