Doc Review Assist

Doc review is the AI surface where the cost of being wrong is bounded but persistent. An AI summary that misstates a contract clause is not catastrophic — the junior attorney reviewing the document will catch it — but if the summary is consistently off by 10–15% in subtle ways, the firm spends junior-attorney time correcting AI outputs rather than reading documents directly. That negative-leverage outcome is what kills most law-firm AI deployments in 2026.

Our doc-review-assist surface is designed to avoid that failure mode through three deliberate choices. First, every AI summary carries clause-level citation hover-cards: hover any sentence in the summary and the source clause is highlighted in the source document with paragraph-and-line precision. Second, the surface is positioned as a senior-attorney time compressor, not a junior-attorney replacement — the junior still reads the document; the senior reads the AI summary plus the junior's notes. Third, we ship with a per-document-type eval framework (M&A asset purchase agreements, real estate purchase agreements, employment agreements, supply contracts, etc.) and never claim performance on document types we have not specifically tuned for.

The Casetrack retrospective documents what we learned about citation UX the hard way — clause-level granularity is the difference between attorneys trusting the AI summary and attorneys reading the AI summary as 'making things up.' That granularity is non-negotiable in our deployments.