Should you build or buy AI Legal Assistant?
An AI legal assistant is a chatbot or agent trained or grounded on a firm's own precedents, playbooks, and matter files, used for legal research, document drafting, contract review, and summarization. It differs from general AI tools in being tuned to a specific firm's practice areas, house style, and client work.
Build the assistant on a frontier model over your own precedents and workflows, the way Goodwin, White & Case and Morgan & Morgan did, and buy research content (Westlaw, Lexis) as a grounding source. A legal-specific vendor makes sense for firms without an engineering function or for practice areas where citator verification is the whole job.
Build it, buy it, or bridge?
When building makes sense
Building is the stronger case here, and law firms are proving it in production. Dentons launched fleetAI, a GPT-4 chatbot built with Microsoft, in August 2023 for legal research and document analysis. Gunderson Dettmer followed weeks later with ChatGD, its own generative AI chat app running on a secure Microsoft Azure instance of OpenAI's models with retrieval-augmented generation. A&O Shearman went further and built ContractMatrix with Harvey and Microsoft; about 2,000 of the firm's lawyers now use it daily, and A&O reports it cuts contract review time by roughly 30 percent. Morgan & Morgan has put roughly $300 million into its own MX2 platform, with nearly 5,000 monthly users, on the way to a planned $1 billion investment over ten years. The work is grounded in a firm's own precedents and matter files, which a general AI tool cannot supply out of the box. The one piece that stays bought: licensed case law and citator verification, the infrastructure behind legal research, which a firm's own model cannot replicate.
When buying makes sense
Buying still earns its place for the citation and verification layer underneath any assistant. Licensed case-law databases and citators represent decades of curated content and case-law tracking that no firm, however AI-forward, is going to replicate on its own model. Harvey.ai reports more than 2,400 legal organizations building on its platform, evidence that even firms running their own assistants still license research infrastructure alongside them. Buying also makes sense for a firm without the innovation staff to treat an AI tool as an ongoing product: prompt tuning, security review, and hallucination monitoring are real ongoing costs a vendor absorbs. Gunderson Dettmer's own tech stack for ChatGD leans heavily on open-source components it maintains itself, which is a real commitment few firms are staffed to sustain, and it is the clearest sign of how much ongoing engineering a firm-built assistant actually requires once it ships.
The desk read
Build-versus-buy analysis for AI Legal Assistant is being written. In the meantime, the framework that drives every B4 call is on the B4 Index page.
Frequently asked
What is an AI legal assistant?
It's an AI system, usually built on a frontier model like GPT-4 or Claude, that works over a firm's own precedents, playbooks and matter files instead of generic training data. Some firms buy one from a vendor; a growing number, including Dentons and Gunderson Dettmer, built their own instead. The one thing almost nobody builds is the case-law research and citation-checking underneath it, which stays licensed.
Which firms have built their own legal AI assistant?
Beyond Dentons and Gunderson Dettmer, Goodwin Procter built a venture-financing tool on a customized version of Claude with six associates, backed by a planned $25 million a year for further AI products. White & Case launched Atlas, a multi-model assistant on Microsoft's Azure AI Foundry, in late 2025, while also partnering with Legora as a licensed platform. Morgan & Morgan has committed roughly $300 million so far to its own MX2 platform.
Why doesn't case-law research get built too?
Licensed case law and citator verification represent decades of curated legal content and citation tracking. That is a licensing and content problem, not a model problem, so it stays bought even at firms running their own drafting and research assistants.
How much does A&O Shearman's ContractMatrix actually save?
A&O Shearman reports that attorneys using ContractMatrix save an estimated 7 hours on average per contract review, cutting overall review time by roughly 30 percent, with a tracked-changes workflow that keeps a human reviewer in the loop on every proposed edit.
What are the main AI legal assistant vendors?
Representative vendors include Legora, Lexis+ AI, Thomson Reuters CoCounsel. Harvey.ai, used by more than 2, 400 legal organizations, is the most widely adopted platform outside firms' own builds. B4 Pro includes the category score and the full vendor list.