Legal & Compliance

Deal terms / intake form → first draft from template → redlined versions → issues list → executed agreement → obligations summary in CLM

• Draft
  First draft from playbook — AI assembles the draft from the approved template, deal-specific terms, and the clause library · high leverage and playbook-verifiable, but the lawyer owns what the company proposes to be bound by
• Draft
  Counterparty redline review — AI diffs the markup against the playbook, classifies each change as standard/negotiable/escalate, and cites the playbook rule · verifiable against an explicit playbook, but a missed non-standard indemnity is a high-stakes, hard-to-reverse miss
• Draft
  Issues list & negotiation prep — AI drafts the internal summary of open points with fallback positions from prior deals · grounded in deal history, but choosing which hill to die on is the negotiator's call
• Assist
  Negotiation itself — AI supports live with clause alternatives and precedent language · pure relationships-and-leverage territory; the counterparty is a human reading another human
• Automate
  Execution & signature workflow — routing, signature blocks, countersignature chase, filing · procedural, checkable, and every step leaves a system record
• Automate
  Post-signature obligation extraction — AI extracts renewal dates, price escalators, SLAs, and termination windows into the CLM · extraction against source text is verifiable and this is where contracts go to be forgotten

Inbound request/document → intake classification → playbook comparison → auto-approved or redlined version → signed NDA → repository entry

• Automate
  Intake & routing — AI classifies the request (our paper vs. theirs, NDA vs. disguised something-else) and routes accordingly · high-volume classification with cheap human recovery when routing is wrong
• Automate
  Playbook conformance check — AI screens the counterparty's paper clause-by-clause against the NDA playbook and marks pass/fail per position · this is the canonical bounded legal task: explicit rules, low stakes per document, fully verifiable
• Draft
  Standard redline generation — AI applies pre-approved fallback language for each failed position · edits come from an approved library, but a paralegal should eyeball before it leaves the building under the company's name
• Automate*
  Escalation of non-standard terms — AI flags the 10% with unusual provisions (non-competes hiding in NDAs, IP assignment creep) for attorney review (Automate for the flagging) · the whole design goal is that AI over-flags and attorneys only see true exceptions
• Automate
  Repository & obligation tracking — AI files the executed NDA with extracted term, scope, and expiry metadata · verifiable extraction, and unsearchable NDAs are a recurring diligence fire drill

Legal question → research plan → authority collection (cases/statutes/regs) → synthesis of positions → research memo with recommendation

• Draft
  Issue spotting & research planning — AI decomposes the fact pattern into researchable questions and identifies likely bodies of law · strong at breadth, but a missed issue never gets researched, so the attorney owns the frame
• Draft
  Authority retrieval — AI-native research tools find cases, statutes, and regulatory guidance with citations and treatment flags · powerful but never Automate: hallucinated or mis-characterized citations carry sanctions risk, so every cited authority gets human verification against the primary source
• Draft
  Reading & synthesizing authorities — AI summarizes holdings, distinguishes facts, and maps the circuit split · verifiable against the opinions themselves, and the highest-leverage stage — it collapses days of reading to hours
• Assist
  Applying law to facts — the actual legal analysis: how does this authority bear on our client's situation · this is the licensed judgment being paid for; AI proposes analogies, attorney reasons
• Draft
  Memo drafting — AI produces the memo in house format from the attorney's analysis and verified authorities · structure and prose are AI-strong; conclusions and confidence levels are the attorney's
• Assist*
  Cite-checking & final verification — every citation confirmed against the primary source before the memo circulates (Assist only) · post-*Mata v. Avianca* discipline: AI can pre-check quotes and pin cites, but a human confirms, because this stage exists to catch AI

Regulatory change feed → applicability assessment → gap analysis → updated policy → training module & attestations → monitoring reports / issue log

• Automate
  Regulatory change monitoring — AI watches rulemaking dockets, enforcement actions, and guidance across jurisdictions and summarizes what changed · high-volume reading with citations to primary sources; missing something is the status quo it replaces
• Assist
  Applicability & impact assessment — does this rule apply to us, and what breaks? · requires knowing the business's actual activities and licenses; regulatory exposure makes a wrong "doesn't apply" costly
• Draft
  Policy drafting & updating — AI redlines existing policies against the new requirement and drafts new sections · verifiable against the reg text, but policies are what regulators hold the company to verbatim
• Draft
  Training content & delivery — AI generates role-specific scenarios, quizzes, and refresher modules from approved policy · low stakes per artifact and grounded in approved source, but compliance training that misstates policy is discoverable evidence
• Automate*
  Transaction/communication surveillance — AI screens expense reports, trades, or communications against compliance rules and escalates anomalies (Automate for screening) · exactly the high-volume pattern-detection AI is for; disposition of every alert stays human
• Avoid
  Regulatory filings & certifications — the compliance officer signs the SAR, the annual certification, the exam response · personal liability regimes (e.g., SMCR-style, BSA officer liability) make the signature non-delegable by design

Litigation hold notice → collected corpus → processed/deduplicated set → responsiveness & privilege review → production set → deposition/trial preparation materials

• Automate
  Hold notices & custodian tracking — AI drafts hold notices, tracks acknowledgments, and flags lapsed custodians · procedural with a full audit trail; spoliation risk comes from *not* systematizing this
• Draft
  First-pass responsiveness review — AI review (successor to TAR) classifies documents for responsiveness with attorney-validated sampling · court-accepted when validated, and at millions of documents human-only review is the fiction — but the protocol and sampling must be attorney-owned and defensible
• Assist*
  Privilege review — AI flags likely-privileged documents for attorney determination (Assist only) · producing one privileged document can waive privilege broadly; stakes and irreversibility cap AI at pre-sorting
• Draft
  Deposition & witness preparation — AI builds witness-specific chronologies and pulls every document touching a witness from the corpus · verifiable against the record and enormously time-saving; strategy of the examination is counsel's
• Draft
  Case chronology & fact development — AI constructs the master timeline linking documents, testimony, and events · grounded in the record and checkable, but which facts matter is theory-of-the-case judgment
• Draft
  Production & privilege log — AI generates the log entries and validates production format compliance · templated but court-facing; sanctions attach to errors

Invention disclosure → prior-art search report → patent application → office-action responses → granted patent → annuity/renewal decisions → enforcement watch reports

• Draft
  Invention disclosure intake — AI structures inventor submissions, flags prior internal disclosures, and drafts clarifying questions · low stakes and improves inventor throughput, but what's actually novel needs attorney reading
• Draft
  Prior-art searching — AI runs semantic search across patent databases and technical literature, clustering and ranking results · dramatically better recall than keyword search, but a missed reference surfaces later as an invalidity problem, so counsel owns the search conclusion
• Assist
  Application drafting — AI drafts specification sections and claim variants from the disclosure and search results · claim language is the asset itself; scope decisions are irreversible after filing, and one imprecise word narrows twenty years of protection
• Draft
  Office-action response prep — AI summarizes examiner rejections, maps cited art against claims, and drafts response shells · analysis is verifiable against the file wrapper; arguments made become prosecution-history estoppel, so counsel owns them
• Automate*
  Docketing & annuity management — AI tracks worldwide deadlines and drafts keep/abandon recommendations with product-mapping and citation data (Automate for deadlines, Assist for abandon decisions) · deadlines are rule-based with fatal misses, so systematize; abandonment is irreversible strategy
• Draft
  Competitive & infringement watch — AI monitors competitor filings and products, drafting claim-chart starting points · surveillance is high-volume and AI-suited, but asserting infringement is a decision with countersuit consequences