A directory of 26 publicly reported legal workflows built on semi-autonomous AI agents and in-house systems, showing each workflow’s practice area, agent authority, and how automated and human evaluation validate performance.
How to read this directory
Agent
An AI system that carries out a multi-step legal task with some autonomy (it uses tools, reads or writes files, or takes actions), not just a single chat Q&A. Only workflows meeting this scope rule are listed.
Agent type
General-purpose assistant: a broad assistant (like Grok Bot, OpenClaw, or Hermes Agent) configured for a legal task. Specialized legal workflow: a narrow agent built for one legal workflow.
Practice area
The legal domain of the workflow.
Authority of the agent
Read-only = the agent only reads, extracts, or analyzes. Draft-only = it prepares drafts or redlines, while a person sends, files, or incorporates them. Writes to systems = it creates, edits, schedules, or updates records in external tools.
Evaluation
How effective performance of the agent is validated across Automated evaluation (programmatic rules, contradiction checks, sync health checks, agent-to-agent critique, JSON schema validation, or feedback loops) and Human evaluation (where a person reviews, approves, or cite-checks). Parenthetical notes indicate whether a check is (stated) in the source or (inferred) by the editor.