Workflow brittleness in legal AI

Diving deeper into

Healthcare company associate GC on where legal AI products break down

Interview
The happy path works great. Anything else grinds to a halt.
Analyzed 5 sources

This is the core product gap in legal AI, demos prove the model can answer a clean prompt, but real buying decisions depend on whether the software can recover when a business user submits the wrong form, a counterparty changes terms midstream, or an approval route breaks. In lean in-house teams, that failure does not create a small annoyance, it pulls a lawyer back into manual triage and wipes out most of the promised time savings.

  • The interview makes clear that the brittle point is usually workflow, not raw generation quality. The legal team can already get drafting and research from enterprise ChatGPT and Westlaw. What it still lacks is a contract system that preserves context, flags deviations from internal playbooks, and guides non-legal users back on track without legal stepping in.
  • This helps explain why CLM vendors win demos but disappoint after rollout. Luminance sells document review and playbook building, yet the lived complaint here is heavy setup, slow support, and breakdowns when users go off script. Ironclad is larger because it built around the full contract workflow, not just clause analysis, but even there the category risk is that AI review becomes easy to copy while workflow reliability stays hard.
  • Law firms describe the same pattern from another angle. Large firms pilot tools for months, buy small seat counts, and favor products that fit a specific practice or workflow instead of trying to do everything. That is why general platforms like Harvey and Legora can get attention, but stickiness still comes from narrow, repeatable jobs with strong onboarding and user training.

The next winners in legal AI will look less like chat products and more like fault tolerant workflow software. The bar is not writing a clever first draft, it is keeping a contract request moving when the process gets messy, while saving every detail and telling the next person exactly what to do. That is where budget, trust, and real expansion will concentrate.