Everyone knows Harvey exists. Holland & Knight will pay $163-245K if you can operate Harvey, Legora, and Microsoft Copilot together in a real matter. Harvey itself hires legal-AI enablement roles in NYC and SF. Almost nobody has published the actual workflows — the clause review sequences, the red-lining prompts, the matter intake automation, the discovery triage patterns — that in-house counsel and mid-law associates run daily.
This hub is for the in-house counsel, mid-level associate, legal ops manager, or contracts manager who has been told "get good at AI legal tools this quarter." It curates skills on reviewing an MSA against your playbook using Claude or Harvey, red-lining a vendor NDA in under fifteen minutes, running matter intake and conflict checks with Copilot, triaging discovery with Everlaw AI or Kira, and building a firm-wide prompt library that survives partner turnover. Every skill is bar-adjacent and privilege-aware.
Who this hub is for
In-house counsel, AmLaw 200 associates (Y2-Y5), legal ops managers, contracts managers, and legal-tech PMs.
Contract review against a playbook, first-pass red-lining, matter intake and conflict checking, discovery document triage, deposition summary drafting, memo research (with citation verification), timekeeping narrative drafting, and client update emails. What you cannot do yet: unsupervised advice to clients, sign-off on final court filings, and anything requiring fresh case-law citations without human verification.
Harvey wins for firm-wide deployments with matter-context integration and audit-grade privilege handling. Copilot wins for individual attorneys already inside the Microsoft ecosystem (Word, Outlook, Teams). Claude wins for long-context work (reading a 300-page merger agreement in one pass) and for teams that want the smartest model with the fewest guardrails. Most sophisticated firms use two or three.
Use only enterprise-tier AI (ChatGPT Enterprise, Claude for Enterprise, Harvey, Copilot with data residency) that contractually excludes your inputs from training. Enable audit logs. Never paste privileged content into a consumer AI. Configure data residency for the jurisdiction. Have your GC review the AI vendor DPA specifically for training-data exclusion and retention.
Legal engineering is the discipline of building supervised AI agents for legal-domain tasks — think Norm AI's work with $30T AUM clients. It sits between traditional law and software engineering. You do not need to code, but you do need to design prompts, define policies, and supervise agent behavior against a legal rubric. Fast-growing career path for lawyers who like structure.
Load your firm's playbook (approved language, non-negotiables, fallback positions) into a prompt or a Harvey/Claude project. Paste the incoming contract or vendor NDA. Ask for a section-by-section comparison flagging deviations from your playbook, with proposed replacement language. Review every flagged section manually — AI is a first-pass tool, not a final decision maker.
Yes, with transparency. ABA Formal Opinion 512 (2024) says billing is fine as long as you bill for time spent, not for AI compute or a fake human-hours equivalent. If AI reduced a task from 4 hours to 30 minutes, bill 30 minutes plus your review time — do not bill 4 hours. Some clients now require an AI disclosure on invoices. Ask.