Three lawyers put legal AI through two weeks of real-world testing
An independent video review of workflows on real matters — what transactional, litigation, and IP lawyers found useful, and where they saw limits.
Discuss this post in AI
Send a pre-filled prompt to ChatGPT, Claude, Gemini, or Perplexity — get a summary, ask follow-ups, or compare ideas from this guide.
Legal AI adviser Liam Barnes gave three lawyers working across transactional, litigation, and intellectual-property matters a legal AI platform and asked them to use it on real work for two weeks. His video brings those observations together with a practical tour of the product.
The result is useful precisely because it is not a product launch or a benchmark we commissioned. It shows where independent users found immediate value, which trade-offs mattered in practice, and where the platform still fell short at the time of testing. We read it against Legal AI from Pattern Automation — the same surfaces: assistant, projects, tabular review, workflows.
What the review covers
The review covers setup, projects, reusable workflows, contract review, multi-document analysis, deployment choices, operating costs, and a feature comparison with commercial legal AI platforms.
Clarification on legal research
Legal AI from Pattern Automation supports U.S. legal research through integration with CourtListener’s case-law database. The assistant can search case law, retrieve and read opinions, locate relevant passages, verify reporter citations, and return linked case references. Characterizing case-law research as a gap does not reflect current capabilities.
What the testing surfaced
The lawyers found practical value in NDA and contract review, while the demonstration of multi-document analysis showed how a platform can structure a larger evidence set into a reviewable table. The video also walks through projects and pre-built workflows rather than assessing the product only through isolated prompts.
The review is equally direct about limitations. It discusses setup and model costs, gaps raised by litigation and IP use cases, and its view of the legal-research capabilities available at the time. As clarified above, Legal AI from Pattern Automation supports U.S. case-law research through CourtListener. The remaining findings are best read as a dated, hands-on snapshot rather than a statement of the current feature set.
Why independent demos matter
Self-hosting and inspectable workflows make a legal stack adaptable, but they do not eliminate model usage, infrastructure, implementation, or professional review. Seeing lawyers work through those trade-offs provides a more useful signal than a feature list alone.
No single evaluation can predict performance for every firm or matter. The strongest takeaway is that teams can now test a substantive legal AI workflow themselves, examine how it works, and compare the result against both their existing process and commercial alternatives.
That is the test we want for Neuro Legal Stuff.