As a lawyer building an AI-powered law firm, I want to like legal AI tools. I should like them. But most of the time, I end up feeling... annoyed.

Every few weeks, I test another shiny new platform hoping it’ll inspire our own roadmap or maybe just make my day-to-day work a bit easier. But almost every time, I walk away unimpressed.

The problem? These tools just aren’t built for how lawyers actually draft and review documents.

They’re often great demos; clever, fast, even a bit magical on the surface. But the moment you try to use them in a real legal workflow, the cracks start to show. These tools aren’t extensions of how we think. They’re trying to be “the brains” of the operation and in doing so, they remove the control, oversight, and professional judgment that define what we do.

HarveyAI

Everyone's talking about it. In reality, it’s a sleek ChatGPT wrapper aimed at big law firms, with heavy enterprise onboarding and support. The Vault, a 10,000-document retrieval system for custom knowledge is probably its best feature. But even that isn’t game-changing; if you’ve got a decent dev team, you can build something similar in-house.

LawInsider

I’ll admit I like their precedent database. It’s wide-ranging and reasonably priced. But their Microsoft Word plugin, which is supposed to bring AI into your drafting environment, just doesn’t deliver. The UX is clunky, the suggestions don’t feel contextual, and overall, it slows things down more than it helps.

Genie AI

In theory, it’s a collaborative platform built around open-sourced, community-written templates. But when those templates aren’t updated, curated, or reviewed for credibility, the result is more DIY headache than helpful shortcut.

LexisNexis

Still one of the most trusted names in law, and for good reason. But their AI offering? It's just RAG (retrieval augmented generation) dressed up in enterprise branding. It searches their knowledge base and summarises results. That’s useful but it’s not new, and it’s not AI in the way people want it to be.

The Common Thread

All of these tools share the same frustrating trait: they try to be too smart, and in doing so, they take away control. They want to replace thought, rather than enhance it.

What I need, what I think most lawyers need, is not a tool that acts like an overconfident trainee. I don’t want a black box that generates a contract and expects me to trust it. I want something that helps me do my job faster, better, and with more confidence without taking the wheel away.

What the "Horseless Carriage" Analogy Gets Right

I recently read “Horseless Carriages” by Pete Koomen, and the comparison really landed. Koomen argues that a lot of AI today is designed to upgrade old workflows, not rethink them. Like the first cars that looked like carriages because people couldn’t imagine anything different.

Legal AI is stuck in that same loop.

It’s trying to automate drafting or reviewing as if we’re just filling in templates or regurgitating precedent. But legal work isn’t that simple. Context matters. Judgment matters. The role of the lawyer isn’t just moving text around, it’s knowing why and when things should change, and what risk that introduces.

We're being asked for user input; write a prompt, click to generate, hope for the best. But what we really need is system input: context-aware tools that understand what we’re working on, see what stage we're at, and offer meaningful, structured support.