Harvey Adds Natural-Language Agent Actions to Review Tables

Harvey has made every action in its Review Tables product executable through plain-language instructions to its AI agent, allowing litigation and deal teams to restructure, correct, and route document review tables without rebuilding them. The feature, disclosed by the company on September 17, 2026, follows the August release of Harvey II, which expanded the agent's capacity for multi-step work across the platform.

Review Tables launched as a way to extract structured answers from large document sets — privilege calls, contract terms, custodian data — but the table itself stayed rigid once populated. Productions arrive in waves, privilege designations change, and partners reframe questions late in the day. Each change previously meant rebuilding columns or starting a new chat thread. Harvey's update attacks that friction directly, letting users revise a column's prompt, output format, or underlying model in place rather than recreating it.

In the workflow Harvey describes, a litigation team converts a free-text privilege column into a fixed classification of Attorney-Client, Work Product, or Not Privileged, attaches a Relativity export to pull custodian metadata in as new columns, and routes flagged privilege rows to a named reviewer in red — all from a single chat. The same mechanism handles bulk cell edits across hundreds of rows, row additions as documents arrive, and grouping of email threads with attachments or complaints with exhibits. Users can also retrieve a table by describing it — "find my most recent review table" — instead of navigating to it, and can move from table to memo or visualization without switching threads. The agent inherits the user's permissions, so existing Vault controls and ethical walls remain in force. Harvey targets M&A diligence and litigation discovery teams as the primary beneficiaries.

Industry Implications

The update pushes Harvey deeper into the document-review layer long dominated by e-discovery incumbents such as Relativity and Everlaw, and it does so from the workflow side rather than the processing side. By letting customers pull Relativity exports into Harvey tables, the company positions itself as the orchestration layer above existing review infrastructure, not a replacement for it. That is a lower-resistance sales motion — but it also raises the stakes for Thomson Reuters, which has been layering agentic features onto CoCounsel, and for standalone review vendors that risk becoming data sources rather than destinations. For legal operations leaders, the practical shift is that table maintenance, historically a paralegal and project-management cost, becomes a prompt.

DreamLegal Perspective

The strategic signal here is less about table editing than about where Harvey wants to sit in the stack. Agent actions inside a structured review artifact move the product from answer generation toward process ownership, and the Relativity import is a deliberate interoperability play that avoids a head-on collision with e-discovery vendors while quietly making Harvey the interface where review decisions get made. Legal ops teams evaluating the category should press vendors on two things: whether agent actions are auditable and permission-scoped at the cell level, and whether outputs survive export back into incumbent review platforms without loss. Buyers who cannot answer both will inherit the same rework burden in a new interface.