Legal AI Tools Compared: A Decision Framework for Disputes Practices

A year ago, a disputes practice head evaluating legal AI tools had two real options. Today, the same partner faces Claude Cowork, Claude for Word, Harvey, Legora, legal plugins, skills, connectors, and a vocabulary that barely existed twelve months ago. We analysed Claude Cowork's legal plugin in February, when the announcement triggered an 18 percent drop in Thomson Reuters stock and a $285 billion selloff across legal technology shares (Morningstar). Since then, the landscape has only accelerated.

The problem is no longer adoption. It is navigation. Most law firms do not have dedicated innovation departments. The partner running a disputes practice is also the person expected to evaluate AI legal tools, compare vendors, decode the terminology, and make a decision that will affect the team for years. We hear this from practice heads across the UK, France, and the Middle East: "Every vendor sounds the same. I do not know what half of these terms mean. And I do not have time to find out."

This blog is designed to fix that. We map out exactly what each major legal AI tool does, in plain language. We compare them against the actual workflow stages of a disputes practice. And we provide a clear framework for choosing, informed by what we have heard directly from the firms evaluating these tools. If you read nothing else, you will leave with three questions that cut through the noise faster than any vendor demo.

Legal AI Tools Explained: What Each Platform Actually Does

Claude Cowork and the Legal Plugin

Claude Cowork is Anthropic's desktop agent application, launched in January 2026. It operates at the file-system level, reading and writing files on a lawyer's local machine. What makes it distinct is its layered architecture:

• Skills are markdown files that teach Claude a repeatable workflow, essentially encoding a firm's specific way of handling a task.
• Plugins are bundled packages containing multiple skills and slash commands. The Legal Plugin provides structured commands like /legal:review-contract for clause-by-clause analysis, /legal:triage-nda for rapid pre-screening, and /legal:compliance-check for regulatory framework verification.
• Connectors use the Model Context Protocol (MCP), an open standard, to link Claude to external systems like iManage, Box, or Egnyte.

Claude for Word

Claude for Word, released as a public beta on April 10, 2026, is a native Word sidebar available through Microsoft AppSource. It places Claude directly inside the document environment. Edits land as native tracked changes. Claude reads comment threads, edits the anchored text, and replies explaining what it changed. It preserves multi-level legal numbering, defined terms, and styles. It also offers counterparty analysis, identifying which revisions from opposing counsel are dealbreakers. This is available to Claude Team and Enterprise subscribers.

Harvey

Harvey is the enterprise AI platform targeting BigLaw. Its primary competitive advantage is a partnership with LexisNexis, providing legal research outputs grounded in authoritative case law and statutes. It is designed around team-based diligence workflows and collaborative reasoning.

From a pricing perspective, Harvey operates at the top end of the market. Reports from firms we have spoken with indicate pricing upwards of $1,000 per user per month, with mandatory minimum seat counts and inflexible annual contracts. For a 20-person disputes team, that can mean a six-figure annual commitment before the tool has been tested in a live matter. Several practice heads at midsize firms have told us they reached out to Harvey but either received no response or were told their firm was not large enough to be considered. This is a pattern worth noting: the firms that arguably need AI for law firms the most to scale their caseload without overhiring are sometimes the ones that enterprise vendors deprioritise.

Legora

Legora (formerly Leya) focuses on large-scale portfolio review and tabular extraction. It is strongest in M&A diligence and mass extraction, where the goal is to pull specific data points from thousands of agreements into a structured view. Its Workflows feature, launched in March 2026, adds a multi-step agentic framework for automating sequences like compliance checks and chronology creation. Legora's pricing is negotiated on a per-firm basis, but market reports and conversations with prospective users suggest it sits in a similarly premium bracket, oriented toward larger firms with dedicated budgets for legal technology.

Where Each Legal AI Tool Fits in a Disputes Workflow

The clearest way to evaluate legal AI software is to map it against the five stages of a disputes workflow. No single legal AI tool covers all five stages well. This is the structural reality that most feature comparison grids obscure.

Document Intake and Organisation

A disputes matter often begins with thousands of unstructured files arriving in bulk: scanned PDFs, bundled exhibits, multilingual correspondence. Claude Cowork can process files on a local machine but is limited to what sits in a folder. Harvey uses vault-based enterprise infrastructure. Legora handles structured document sets effectively. None of them offer automated titling, dating, categorisation, or multilingual OCR as a native part of the intake workflow. Kallam was purpose-built for this stage, offering automated structuring and multilingual OCR as native intake features.

Chronology and Case Structure

Building a chronology is the backbone of disputes work. Claude Cowork can be taught this through custom skills, but the firm must build those skills from scratch. Harvey is not primarily a chronology tool. Legora offers tabular extraction but is designed for portfolio analysis, not dispute chronologies where contextual dating and narrative sequencing matter. Kallam generates chronologies automatically from contextual dates during document processing, without requiring custom configuration.

Research and Analysis

Harvey is relevant here, given its LexisNexis integration and team-based reasoning workflows. Legora also offers research capabilities, with access to authoritative legal sources including case law, statutes, and regulations, combined with internal firm knowledge through DMS integrations. Claude Cowork has broad general knowledge but no access to legal databases. Claude for Word is not a research tool. Kallam offers an advanced web search feature with access to publicly available legal information, alongside an AI agent grounded strictly in uploaded documents.

Drafting and Submissions

Claude for Word is the most significant recent development at this stage, resolving a persistent frustration: the need to leave the document to interact with AI. Its tracked changes and formatting preservation make it genuinely useful for redlining. Harvey and Legora both offer Word add-ins as well, but it is worth noting what we have heard from users. Several lawyers who tested these integrations reported that the add-ins were unstable during their evaluations. They found themselves preferring the chat interface for drafting instead. Claude Cowork, operating at the file-system level, also handles drafting workflows well through its skills architecture. Kallam offers an agent drafting mode alongside a Word add-in that manages exhibit numbering and citation insertion directly in the document.

Review and Counterparty Analysis

Claude for Word offers native counterparty analysis directly in the sidebar. Legora excels at portfolio-scale review across large document sets. Harvey provides team-based review workflows. Claude Cowork relies on custom skills for any structured review process. Kallam offers a Virtual Intern that extracts structured data across documents into a tabular view for systematic review.

The pattern is clear. Most legal AI tools excel at one or two stages and are weak or absent at the others. Kallam was designed to address this gap, covering all five stages within a single disputes-focused platform. A practice head who needs to solve a drafting problem will reach different conclusions than one drowning in unstructured document intake.

The Privilege Question: Can You Safely Use AI for Legal Work?

Workflow fit is only half the decision. The other half is whether your chosen AI legal tool can safely handle client-confidential material.

The Heppner Ruling

In February 2026, Judge Jed Rakoff of the Southern District of New York ruled in United States v. Heppner. The court held that documents generated by a criminal defendant using a consumer-grade version of Claude were not protected by attorney-client privilege (Harvard Law Review; Gibson Dunn). The court's reasoning was direct. Claude is "plainly not an attorney," and as Rakoff noted, all "recognised privileges" require "a trusting human relationship." Anthropic's consumer privacy policy at the time permitted the company to collect user inputs and outputs for model training, and to disclose data to third parties including government authorities. The defendant had used the tool independently, without counsel's direction. Rakoff compared the arrangement to "shouting secrets in a crowded public elevator" (Lawfare).

The Path to Privilege Protection

The ruling left one path open. Had counsel directed the defendant to use Claude, the tool might have functioned as a "lawyer's agent" under the Kovel doctrine (United States v. Kovel, 296 F.2d 918, 2d Cir. 1961), potentially preserving privilege. The Kovel doctrine extends attorney-client privilege to non-lawyer third parties when they act as agents of the attorney and their involvement is necessary to assist in providing legal advice. In the AI context, this would require counsel to direct and supervise the tool's use. Enterprise-grade confidentiality safeguards must be in place: zero data retention, dedicated tenancy, and formal documentation of the AI's role as counsel's agent (Orrick). This distinction draws a hard line between consumer-grade AI and enterprise deployments with documented counsel oversight.

The GDPR Gap

For firms operating in Europe, a separate constraint applies. The Claude API has offered EU data residency since August 2025. But claude.ai and Claude Desktop do not yet support EU-only processing. In practice, this means European lawyers must perform a mental GDPR assessment before every prompt involving personal data or client documents. Several EU-sovereign alternatives have emerged to address this gap, including Mistral AI, which secured $830 million in debt financing in March 2026 to build sovereign data infrastructure in France.

The question is not just "which tool fits my workflow." It is "which tool can I safely use for privileged, client-confidential material without risking a waiver."

What We Built for Disputes and Why It Looks Different

The workflow comparison above reveals a structural gap: most tools cover one or two stages well, but no horizontal platform covers the full disputes workflow. We studied these gaps before writing a single line of code. Like most modern AI platforms, Kallam uses foundation models under the hood. But the value is not in the model. It is in the proprietary workflow layer that connects document intake, chronology, search, and drafting into a single integrated pipeline for disputes.

The End-to-End Disputes Workflow

The difference is integration. The features described in each workflow stage above exist within a single platform. A disputes team does not need to switch between tools at each stage or build custom skills from scratch. Documents arrive, get structured, populate a searchable chronology, and flow into drafting with exhibit citations intact.

For bundled or scanned PDFs, a built-in document splitter lets the user split files into individual subdocuments before uploading, so processing applies to each document individually rather than the whole bundle. The AI agent operates under a strict RAG architecture that constrains every output to source material. Every citation is displayed side by side with the original document for immediate verification. When it is time to draft, the Word add-in auto-numbers exhibits by order of appearance and distinguishes factual from legal exhibits.

Compliance and Pricing

The platform is hosted on Azure in the EU. GDPR-compliant. Zero data retention for AI model training. Full per-client data isolation. Prices start at $65 per seat per month.

The Technical Partnership

But the product is only part of it. The firms seeing the strongest results are working with us not as a vendor but as a technical partner. We design the implementation strategy of the tool around each firm's specific practice: how their team structures matters, where their hours actually go, what their submission workflows look like. We take their feedback and ship improvements on top of it. We stay in the room after the contract is signed. A feature can be replicated in a matter of days. That commitment cannot.

How to Choose: Three Questions for Your Disputes Practice

If you are a disputes practice head evaluating legal AI tools, these three questions will cut through the noise faster than any feature matrix.

First, what is the actual bottleneck? If your team spends most of their time on drafting and redlining, Claude Cowork or Claude for Word may be sufficient. If the bottleneck is document intake, structuring thousands of files, and building chronologies, you need a platform that handles bulk processing natively. The answer determines which category of tool you need, not which brand.

Second, what are your privilege, data residency, and budget requirements? If you handle EU-regulated disputes or privileged client material, consumer-grade tools are not an option regardless of capability. Verify zero data retention, hosting jurisdiction, and enterprise-grade access controls before evaluating features. After Heppner, this is not a preference. It is a professional obligation. And be honest about budget. If a platform requires $1,000 per seat with a 50-user minimum and an annual lock-in, it does not matter how good the features are if your firm cannot access them. The market currently has a gap between what enterprise vendors offer and what midsize disputes practices can realistically commit to. Identifying where you sit in that gap is essential before you start comparing feature lists.

Third, do you need a tool or a partner? A tool solves today's problem. A partner designs the implementation strategy around your practice and evolves with your workflow. The firms that have moved furthest with AI for law firms did not pick vendors based on feature checklists alone. They found technical partners who delivered both a strong product and a genuine commitment to making the engagement work.

If you are evaluating AI legal tools for a disputes practice and want to cut through the noise, we are happy to walk through this framework with your team. Explore Kallam AI or reach out for a conversation.