We'd like to think Anthropic was inspired by us. We can't prove it. They probably won't confirm it. It's probably also not true. But when we launched Blueprints, Arna's format for packaging reusable legal processes, and then watched Anthropic release Agent Skills as a general-purpose version of the same concept, we allowed ourselves a quiet moment of pride. Imitation, flattery, and all that. Or perhaps great minds think alike.
It's a sign that the industry is converging on the same insight: AI agents need more than raw intelligence. They need structured, reusable know-how they can pick up and apply consistently so that users can rely on predictable behaviour. Here's how the two approaches compare, and where they part ways.
What are Agent Skills
A Skill is a standardised, packaged set of instructions that an AI agent can pick up and use for a specific task. Instead of explaining from scratch how to do something every time, you write the instructions once, package them up, and any compatible AI tool can load and follow them. The goal is portability and reusability: build a Skill once, use it across different AI products without rebuilding from scratch. It’s a sensible, well-designed standard for general-purpose AI agents.
(A brief note: Anthropic’s product naming in this space — Skills, plugins, connectors, MCP servers has become genuinely hard to follow. We don’t blame them; it happens with fast-growing companies. They are cleaning it up. But it’s worth knowing that the “legal thing” they launched with the most pomp is actually called a plugin, not a Skill.)
What are Blueprints and Why They Go Further
Arna runs on Blueprints: curated packs of expert knowledge, templates, strategies and processes designed to solve specific legal challenges from start to finish. They require the user to provide information specific to the organisation for which a blueprint is used.
And here’s the fundamental distinction: Skills are methods. Blueprints are strategies. A Skill tells an agent how to perform a specific task. A Blueprint runs an entire process from a triggering event through to a completed outcome, calling the right tools at each step, creating deliverables, routing work to the right people, and deciding when a human needs to be involved.
To make this concrete: a Skill for contract review tells an agent how to extract key clauses from a document. A Blueprint for contract review kicks off the moment Arna receives it, she flags deviations against your playbook, and escalates to in-house or outside counsel if a redline threshold is hit, chases approvals, and logs the outcome wherever you instruct her to. The Skill is a technique. The Blueprint is the whole process.
That distinction matters especially in legal work, where no step happens in isolation. A GDPR data subject request isn’t a single task. It’s a sequence of decisions, verifications, drafts, approvals, and deadlines, some of which a machine can handle, and some of which require a qualified human. A Blueprint knows when to keep moving and when to pause.
The Bigger Picture
Agent Skills is building useful infrastructure for the broad arrange of human users. We respect that, and we’re glad the concept has gained traction.
Blueprints are doing something more specific: encoding not just how to do work, but instructing an AI agent how to orchestrate entire processes end-to-end with more or less human intervention. Legal and compliance work carries professional responsibility that general-purpose frameworks simply weren’t built to handle.
Arna’s whole thesis is that legal AI shouldn’t just answer questions. It should run entire processes, moving work forward across multiple people and steps without lawyers becoming the bottleneck at every turn. Blueprints are the building blocks of that. Structured, predictable, repeatable and built for a domain where the stakes are real.
Same concept at the surface. Different ambitions underneath.
A Question for You
Anthropic called theirs Skills. We called ours Blueprints, and we think we both proposed a name that fits its purpose.
But we’re curious what you think.
Does "Blueprints" feel right for our purpose, or would you have suggested a different name? Drop a comment or reach out. We'd genuinely like to know. We want to name things in ways that resonate with the people using them.