Should You Try That AI Legal Tool?
You Do Want AI-Enhanced Legal Workflows. Do You Want This One?
I was hosting an AI panel at an extraordinary legal conference last week and, when I’m around other lawyers thinking about AI, I’m often asked if I’ve tried or heard about a particular “[AI Legal Tool]” on the market. Typically, these tools combine workflow automation or task-specific templates with a large language model (LLM)—the AI core that handles complex language tasks. A broader term for these products is “LLM wrappers”.
I haven’t personally used every AI legal tool, but I do have some practical advice for evaluating them. Before investing resources, first test your workflows directly in one of the advanced frontier AI models available (right now, that essentially means Claude 3.7 Sonnet, Gemini 2.5, or GPT o3 – each is racing the other to be the best and so none stay in the lead for very long).
Have their terms of service vetted before proceeding, to ensure they fit your confidentiality requirements. Then subscribe for the level of protection you need (relatively cheaply compared to the legal tools) and unsubscribe after a few months when you’re done.
Run your real work through it. At a minimum, test the AI on tasks like drafting client communications, workshopping client strategies, summarizing legal materials, unearthing subtext, and preparing and revising agreement clauses.
This gives you a clear benchmark of what the most advanced AI can do – without additional layers, but also without custom benefits.
Only after establishing this baseline should you assess any “[AI Legal Tool].” Consider whether its added features – like ease of use, structured workflows, audit trails, collaboration tools, and ideally jurisdiction-specific legal research – genuinely improve your output. Then weigh those gains against the cost, complexity, and training overhead.
Consider carefully whether these added features justify the “[AI Legal Tool]” costs and complexity. This helps avoid committing too early to a tool that adds little more than polish. Slick interfaces and lofty claims are tempting, but only meaningful improvements to your actual workflow justify the investment.
If you do take this path, I’d be genuinely curious to hear what you learn – especially where a specific tool clearly outperforms the frontier models alone.