Legal expertise has a problem
It's trapped in people's heads. When a senior partner retires, thirty years of pattern recognition walks out the door. When an associate leaves for another firm, their methodology goes with them.
Firms try to solve this with knowledge management systems. They build document repositories, write practice guides, create checklists. But these are passive. They sit there, waiting to be read. And mostly, they aren't.
Junior lawyers learn by osmosis. They watch, they ask questions, they make mistakes, they gradually absorb how things are done. It works, but it's slow and inconsistent. Two associates trained by different partners will review contracts differently.
This is the state of legal knowledge: valuable, fragile, and frustratingly non-transferable.
From knowledge to methodology
AI agents are increasingly capable of legal work. They can read a contract, identify a non-compete clause, summarize its terms. That part is commoditized. Many models can do it.
What no model knows out of the box is that your firm flags non-competes over 12 months as concerning, that you always check for carve-outs protecting general skills, that your partners want findings structured as Key Terms / Risk Areas / Suggested Revisions.
That's methodology. And methodology is what differentiates one firm from another. Agent Skills let you encode it.
What are Agent Skills
A Skill is how you close that gap between generic legal knowledge and how your firm actually practices.
When a lawyer builds a Skill, they're packaging their methodology into something an AI agent can pick up and apply. Every time, consistently.
The format is simple: a folder with instructions, checklists, reference materials. A senior associate who's never written a line of code can create a Skill that captures how they review NDAs, what they flag, how they structure their findings.
What changes in practice
Three things happen when methodology becomes portable.
- Knowledge preservation. When someone leaves, their Skills stay. The expertise they encoded keeps working.
- Repeatable output. Right now, if you ask an AI to review a contract, you'll get a different result each time. Different structure, different emphasis, different things missed. A Skill pins that down.
- Compounding quality. A Skill isn't static. When a review misses something, you update the Skill. When a partner flags a new risk pattern, you add it. Every improvement applies to every future review automatically.
An open standard
Skills aren't locked to one AI provider. The format is an open specification, already supported by major models including Claude, Gemini, and OpenAI's Codex.
This matters. It means the methodology you encode today isn't trapped in one vendor's ecosystem. As AI platforms evolve, your Skills stay portable. You invest in capturing expertise, not in a particular product.
The long-term goal is a common language for teaching AI agents how to do specialized legal work.
The economics shift
If methodology can be encoded, competitive advantage changes.
Small firms can punch above their weight. A three-partner boutique with well-built Skills can deliver consistency that rivals a BigLaw practice.
Routine work commoditizes faster. If an AI can apply a firm's methodology reliably, the value of having a human do routine reviews diminishes. This isn't new (it's been happening for years) but Skills accelerate it.
Value migrates. The premium shifts from "doing the work" to "designing how work gets done." The lawyer becomes a methodology architect, an orchestrator, and a reviewer. The AI does the rest.