Wrappers Aren't Enough. But Neither Is Raw Frontier AI. Law Needs Infrastructure.
The Wrapper Problem
A wrapper takes a general-purpose language model and adds a legal-sounding interface. The model itself has not changed. Its training data is not curated for legal accuracy. Its output format is not structured for citation or audit. Its context window does not persist across a matter. Wrappers are not legal AI — they are legal-colored general AI, and the distinction matters enormously when a brief is filed or a deal is closed.
The Raw Frontier Problem
On the other end, some sophisticated teams have gone directly to frontier model APIs and are building internal tooling on top. The results are often impressive but fragile. The infrastructure burden — fine-tuning, RAG pipelines, hallucination monitoring, jurisdiction-specific validation — falls entirely on internal engineering teams. This is viable for BigLaw with deep technology investment, but it is not a strategy available to the vast majority of the legal market.
What Infrastructure Actually Means
Infrastructure means the hard problems are solved once, correctly, and available to every firm that builds on top. Legal infrastructure means: jurisdiction-aware retrieval that understands citation hierarchies; document processing that handles the actual formats lawyers use; entity extraction tuned to legal concepts rather than general knowledge graph entities; audit logging that satisfies professional responsibility requirements; and output formats that integrate with existing case management and billing systems.
The Durable Advantage
Firms that adopt purpose-built legal infrastructure now are not just getting a productivity tool — they are accumulating institutional data. Every matter processed through the infrastructure trains the firm's private context. Retrieval improves. Templates sharpen. The system learns which arguments have historically worked in which jurisdictions with which judges. This data flywheel is not available to firms running wrappers, and it is not easily replicated by competitors who start later.
Irys as Infrastructure
Irys is not a chat tool with a legal theme. It is an infrastructure layer designed to be the substrate on which legal AI workflows are built. The APIs are designed for integration. The data model is designed for persistence. The retrieval pipeline is designed for legal accuracy. And the output contracts are designed to plug into the systems law firms already depend on — not to replace them.
