← back
Don't Build Agents, Build Skills Instead – Barry Zhang & Mahesh Murag, Anthropic
Takeaway
Stop building bespoke agents per domain — package expertise as portable, file-based 'skills' on top of a universal code-executing agent.
Summary
- Anthropic argues code is the universal interface: with Claude Code as a general-purpose agent, scaffolding shrinks to just bash + filesystem instead of bespoke per-domain tools.
- Introduces Agent Skills: folders of composable procedural knowledge (instructions, scripts, references) that anyone can author, version in Git, share via Drive or zip.
- Skills solve the domain-expertise gap that prompts and tools alone can't — they progressively disclose context (name+description first, then full content) to avoid blowing up the context window.
- Skills can include executable scripts as deterministic tools, work with the agent SDK, and are portable across Claude apps including Claude Code and the API.
agentsclaudeskills
Original description
In the past year, we've seen rapid advancement of model intelligence and convergence on agent scaffolding. But there's still a gap: agents often lack the domain expertise and specialized knowledge needed for real-world work. We think Skills are the solution—a minimal form factor for packaging procedural knowledge that agents can dynamically load. It's a portable, composable approach to giving one agent capabilities across domains. In this talk, we'll share how we built Skills at Anthropic, the network effects we're observing, and where we believe this leads: agents writing their own Skills from experience. Our thesis: equipping agents for real-world work means building reusable expertise. Barry: https://twitter.com/barry_zyj Mahesh: https://twitter.com/MaheshMurag