Enterprise AI assistant. or the platform where legal work actually happens?

Harvey positions itself as a broad enterprise platform for legal and professional services, with product surfaces including Assistant, Vault, Workflows, Knowledge, and Ecosystem. That makes it a serious benchmark in the market, especially for buyers looking for a recognizable enterprise legal AI platform with multiple capabilities under one brand.

Is the goal to add a powerful AI platform to the existing legal stack, or to make the workflow layer itself more unified, more persistent, and more operational? For many firms, those are not the same thing.

A team can have excellent AI answers and still be stuck stitching together matter context, reusable work product, drafting history, and workflow continuity across separate tools and habits. That is not transformation. That is a smarter layer on top of fragmentation.

Irys is built around a different thesis. It is the platform where legal work is supposed to happen and compound: drafting, research, document intelligence, iteration, and matter-native workflows living together instead of being scattered across prompts, files, and disconnected systems.

The architecture behind that vision is not just "call an LLM and hope for the best." It is built around a model-agnostic orchestration layer, multi-agent architecture, knowledge graph engine, enhanced legal embeddings, concept-based processing, and precedent-based learning. with the roadmap extending toward fuller workflow automation over time.

Harvey's integrations are geared toward enterprise deployments and internal knowledge systems. For individual lawyers who spend most of their day in Microsoft Word, there is no publicly available Word-native experience.

Irys ships a free Word Add-In. available today on the Microsoft Marketplace. that brings research, drafting assistance, and cite check directly into the Word sidebar. Lawyers can work inside their existing document workflow without switching tools or contexts.

Harvey is enterprise-only and sales-gated. Pricing is not public. Self-serve does not exist. The AI is licensed by seat, typically with per-query or per-workflow quotas. Your firm's usage gets metered against a budget.

Irys One is $299 per seat per month, transparent on the pricing page, same rate whether you are solo or scaling toward enterprise. Every paid plan includes unlimited matter memory, unlimited AI queries, and unlimited chat length. No document caps, no token limits, no session resets, no gated features by tier. The AI is the same on Irys One and Enterprise.

Harvey reportedly starts around $1,200 to $2,000 per seat per month with a 20-seat minimum, putting the floor near $300,000 per year and Am Law 100 contracts at $5 million or more. Those numbers are reachable for large firms but choose to spend wrapper-margin money on a layer that pays the frontier-lab list rates on every query.

On Claude DIY at the other end of the spectrum, the $20-per-seat sticker is not the price either. Anthropic moved Enterprise to per-token billing in April 2026. Realistic legal all-in lands $300 to $500 per seat per month for an occasional partner, $600 to $1,200 for a typical associate, $1,200 to $2,500 for a litigator or transactional partner doing daily document work, and $2,000 to $4,000+ for an M&A or complex-litigation matter lead. On top of that, the build burden (legal-specific skills, citation verification, matter context, integration maintenance, skill versioning across every model release, IT governance) runs $400,000 to $700,000 per year for a 50-attorney firm and $700,000 to $1.2 million per year for an Am Law 100. Irys analysis throughout, methodology in the linked piece.

There is a structural caveat on Harvey worth understanding before signing. Harvey is a wrapper on third-party foundation models. Every query pays the frontier-lab list rates. That cost base creates pressure on pricing over time. Anthropic just set the precedent across the foundation-model market by moving Enterprise to per-token billing. Harvey has not introduced per-token billing yet, and current contracts often include intro discounts that expire at renewal. The number you sign today is not necessarily the number at year two, the number when you add seats, or the number when you upgrade. Wrapper economics make repricing a question of when, not if.

Irys absorbs that math and is structurally different. Our cost base is not set by a frontier lab. The proprietary infrastructure absorbs roughly 80% of the work before any model is called. That is the structural reason we can offer one number per seat, no token billing, and no usage caps. At enterprise scale Irys pricing is set through sales engagement, but lands materially below Harvey and below Claude DIY all-in at any reasonable market-rate assumption. Annual customers lock in the rate. We are the partner, not an upsell-driven product.

Full breakdown with methodology and sources: irys.ai/blog/real-decision-harvey-legora-claude-irys

The choice is not "which product has AI." Both do. The choice is what role the product is meant to play inside a legal team.

Harvey is a credible enterprise legal AI platform. Irys is making the more aggressive bet that AI should not merely sit beside the workflow. It should become the workflow layer where legal work product accumulates, standards emerge, and teams build operational leverage over time.