This is how I'm seeing the future of the legal workspace: AI agents that do the work your way with Skills + hyper-customized interfaces for human review with VS Code extensions. Here's a concrete example on a feature most are familiar with: Tabular Review.
In the demo below, Claude is given a custom-built Tabular Review skill (link in comment) that lets it perform an extractive analysis of a sample data room.
You see it open the documents, read their content, take notes in temporary files, then save all the extracted information in a structured Excel file.
That's the agent doing the work in a deterministic way, thanks to the methodology described in that custom skill.
Then comes the human review, where interfaces still add real value.
In the demo, I'm using an open-source Excel editor extension. It lets you open the sheets directly in VS Code, filter, search, edit. All you need to properly review the agent's work, without ever needing to leave the app.
And here's what I find really cool: you can ask Claude to build you an interface. Your own small app, living right inside VS Code, that reflects exactly how you think about the task. You decide what it looks like. For once, the tool adapts to you instead of the other way around.
One thing on security though. I'd recommend downloading extensions from organizations you know and trust (ones with a reputation and a face you can hold accountable), just like you would with any software. For the rest, prefer open-source extensions you (or Claude) can inspect. There are many security review skills out there that can perform a deep audit of the code for you. And of course, if you build your own, you control what's in there.
Still early. But the pieces are coming together.