WorkBuddy handles legal briefs and contract reviews, but multi-stage workflows break down. Anyone else experienced this?

Following up on my previous post about material aggregation, I want to push the question further. WorkBuddy gets messy with multi-stage processes for defense statements and contract reviews. Should I blame the tool, or did I fail to break the process into verifiable steps?

I recently had a counter-intuitive feeling: the easiest place for AI drafting legal documents to go wrong is when it arbitrarily supplements facts. Last night, working late until 11 PM, I saw a project for a Chinese legal document assistant that claimed it doesn't rely on template filling but chains case intake analysis, procedural review, evidence analysis, and defense strategy all the way through. I felt tempted to copy their homework. Coincidentally, I had a batch of supplier framework agreements, similar case search materials, and some email attachments, so I tried throwing them into WorkBuddy's workspace. I've used WorkBuddy for about a month and am familiar with single-point functions like document comparison, sentence-by-sentence comparison, and semantic comparison. The workspace was set up just these past few days, contract summarization is newly learned, and expert mode is currently being tested.

This time, I want to ask a more upstream process question. If I break defense statements and contract reviews into multiple stages—first identifying material types, then extracting disputed clauses, then listing evidence gaps, and finally generating modification suggestions—is WorkBuddy suitable for chaining these via workspace, expert mode, and contract summaries? My experience over these past few days suggests that single-document comparisons, sentence-by-sentence comparisons, and semantic comparisons are relatively stable for point-like tasks. However, once multi-source materials are put into one flow, how stages hand off, which outputs are drafts only, and which fields require manual confirmation seem to need extra rules.

Maybe I misunderstood. I initially thought just throwing PDFs and Word docs together would work. Later, I realized the hard part in multi-stage flows is defining the input scope for each step. Which materials go only to procedural review, which only to evidence analysis, and which serve only as attachment indexes. Based on my experience, WorkBuddy is stable for single-document comparisons, but when involving multi-source, multi-stage document flows, I struggle to keep it under control.

But now I'm truly unsure if the process breakdown wasn't detailed enough, or if such multi-stage tasks simply aren't suitable for AI to run end-to-end with one click. Has anyone successfully run defense statements or contract reviews broken into steps within WorkBuddy? Did I misconfigure something, or should I organize materials into a table before feeding them in?