AI Agents for Lawyers

Practice Areas

AI for Corporate Lawyers: Governance, Board Papers and Company Secretarial Work

AI corporate lawyers use models for board minutes, resolutions, entity compliance and shareholder documents, and scheduled agents for housekeeping. Design and gates.

By Dr. Niklas SchmidtUpdated 13 November 20269 min read
On this page
  1. Why housekeeping is the best agent candidate in a firm
  2. Minutes and resolutions: from the pack, not from memory
  3. A corporate-housekeeping agent design
  4. Cap tables, share numbers and the arithmetic
  5. Advising boards on AI oversight
  6. What corporate clients now expect
  7. Setting this up in a corporate team

Anthropic's release notes for Claude Opus 5 (24 July 2026) name corporate governance as one of the two practice areas with the biggest gains, and its open-source corporate plugin ships board consents behind a review gate. Corporate lawyers can use AI today for minutes, resolutions, entity compliance and shareholder documents, and for a scheduled housekeeping agent that drafts routine work into a queue. What the lawyer keeps is every approval, every filing and every number.

Why housekeeping is the best agent candidate in a firm

Gartner's rule of 25 June 2025 is the useful test: use "AI agents when decisions are needed, automation for routine workflows and assistants for simple retrieval", and beware "agent washing", since Gartner counted only about 130 real agentic vendors and predicts more than 40% of agentic projects will be cancelled by the end of 2027. Anthropic's own distinction runs the same way: workflows follow predefined code paths; agents direct their own process. Corporate housekeeping is a workflow. Annual resolutions, register updates, confirmation filings, officer changes, subsidiary-level approvals and minute-book maintenance recur on a calendar, come from templates, and have few ambiguous branches. That is what makes them safe to automate with approval gates, and it is where the distinction in workflow agent and what AI agents are matters in practice.

The vendor evidence points the same way. Anthropic's Opus 5 announcement claims first-turn NDA redlines "nearly double Opus 4.8", a "clear step up in performance on legal agent work" with the biggest gains "in practice areas like corporate governance and arbitration", and financial-modelling accuracy up nine points with a third fewer tool calls. Its release notes add that the model "verifies its own work without being told to", which means old prompts ending "add a final verification step" now over-verify. The open-source Claude for Legal repository's corporate-legal plugin covers tabular diligence with "every cell cited", disclosure schedules, closing checklists, board consents and entity compliance, with a dataroom-watcher agent and a review gate before any board consent executes; it also detects Luminance or Kira output for quality checking. Four of the plugins, including corporate, can run headless as Managed Agents. Details of the plugin set are in Claude for Legal plugins explained.

Minutes and resolutions: from the pack, not from memory

TaskInputs the model needsWhat it producesWhat the lawyer checks
Board minutesAgenda, board pack, recording or CoSec notes, minute templateDraft minutes, each item tied to an agenda point and a sourceEvery resolution's wording; attendance, quorum, conflicts declared; anything the model "remembered" that was not said
Written resolutions and consentsTemplate, the decision, the constitutional documentsDraft resolution with recitals from the packAuthority, thresholds, signatories, dating; the plugin's review gate is the model for this
Shareholder documentsArticles or bylaws, shareholders' agreement, cap tableNotices, proxies, circulars from templatesNotice periods, majorities, pre-emption mechanics
Entity complianceRegister extract, filing calendar, last year's filingsDeadline list, draft filings, register updatesEvery date and every filing against the register
Regulatory filingsThe form, the source data, prior filingsPopulated draft with source referencesEvery field against source; the form's current version

Two rules keep minutes honest. The model drafts from the pack and the recording of the meeting, never from what a board "usually" resolves, and every sentence in the draft that is not traceable to an input is removed. And recording comes with consent: the New York City Bar's Formal Opinion 2025-6 (22 December 2025) and Formal Opinion 2026-2 (5 August 2026) require consent of all parties and treat not recording as the default, and a board meeting has non-client participants. The setup is in AI meeting notes for lawyers.

You are assisting a corporate lawyer with draft board minutes.
Work only from the uploaded agenda, board pack, and meeting transcript
or notes, and the uploaded minute template.

For each agenda item, draft: (a) the matters considered, in one to three
sentences, citing the pack page or transcript timestamp; (b) any
resolution passed, in the template's form, quoting the wording used in
the meeting; (c) declarations of interest and abstentions as recorded.

Record attendance, apologies, quorum and the chair as stated in the
transcript. Where the transcript is unclear or a resolution's wording is
incomplete, insert "[CHECK: ...]" with the timestamp. Do not add
resolutions, approvals or discussion that do not appear in the inputs.

A corporate-housekeeping agent design

This is the design that survives a partner's questions. It is deliberately a workflow with gates, not an autonomous agent.

ElementSpecification
TriggerScheduled (daily at 06:00) plus manual run; never triggered by inbound email
InputsEntity register (read-only export or API); DMS folders for the entity group via a permission-bound connector; the firm's template library; the filing calendar
ConnectorDMS access through the iManage MCP server (14 May 2026; permission-bound, logged, respects ethical walls) or NetDocuments' ndMAX MCP, which runs under the user's own credentials so "documents never leave the platform" and IT can switch off with one toggle
Steps1. Read the calendar and register; list items due in the next 60 days. 2. For each item, locate last year's document and the current template. 3. Draft the new document with every variable (names, dates, share numbers, thresholds) taken from the register and flagged with its source. 4. Compare against last year's version and list differences. 5. Place the draft and the diff in a review queue. 6. Write a run log
GatesNo filing, sending, e-signature or register write of any kind. A lawyer approves each queued item. Anything with a variable the agent could not source is queued as "blocked", not drafted
OutputsDrafts in the DMS under a "draft-agent" folder; a morning summary to the responsible lawyer; a log with every document read and written
RuntimeClaude Managed Agents (public beta 8 April 2026; sandboxing, credential vault, $0.08 per session-hour plus tokens) or a self-hosted n8n workflow calling a business-tier model
Kill switchThe connector's admin toggle; the agent's credentials can be revoked independently of the lawyer's
Review cadenceWeekly sample of approved items against source for the first quarter; the log retained for the file

The pattern is not exotic. Simmons & Simmons reported "hundreds of agents now in operation" (1 July 2026); Legora's aOS pitch is counterparty redlines "reviewed, flagged and responded to by the time lawyers arrive" (7 May 2026); a three-lawyer German firm runs its intake and deadline capture on a self-hosted n8n server for €20–30 a month with a human in the loop for every deadline (case study). The DMS connection is the step most firms get wrong, and it is described in connecting Claude to iManage or NetDocuments with MCP; the general build path is in n8n for law firms. Module 4 of the AI Lab for Lawyers course builds a scheduled agent with exactly this queue-and-approve structure in n8n, without code.

Cap tables, share numbers and the arithmetic

Corporate work is full of numbers that must be exactly right: share counts, percentages after a round, option-pool dilution, thresholds in a shareholders' agreement. Models do arithmetic badly across many inputs and confidently in every case. Anthropic's nine-point improvement in financial-modelling accuracy is welcome and is not a reason to let a model compute a cap table. The working rule is that a spreadsheet or a purpose-built tool produces the figures and the model drafts around them, citing the cell.

The tools a corporate associate can build for this are small and cheap. Jamie Tso, a senior associate at Clifford Chance, has described building a track-changes analyser, a side-letter tabulator, a signature-page extractor and an ownership-chart generator for roughly nothing on Google AI Studio, with synthetic data and never client data in the prototype. A signature-page sorter or an ownership-chart generator is the kind of utility a corporate team will use every week, and vibe coding has made it a two-hour job. For deal-stage work, diligence grids and closing checklists, the M&A page covers the agents and the numbers.

Advising boards on AI oversight

Corporate lawyers are being asked the AI question from the other side of the table. Deloitte's survey of 121 senior legal leaders (April–May 2026) found 61% piloting agentic AI and only 2% with no AI adoption; boards want to know what that means for them. Clifford Chance's briefing "Agentic AI: the liability gap your contracts may not cover" (9 February 2026) sets out the questions: authority of an agent to commit the company, hallucinated commitments, indemnities and insurance. The EU timetable as of September 2026 is that Annex III high-risk obligations were deferred to 2 December 2027 under the Digital Omnibus (Gibson Dunn's summary), while the Article 50 transparency duties and the AI Office's enforcement powers over general-purpose models applied from 2 August 2026 and the Article 4 literacy duty has applied since February 2025. The full map is in the EU AI Act for lawyers and the Omnibus changes in the Digital Omnibus on AI.

Anthropic's ai-governance-legal plugin is a serviceable starting kit for the paperwork: an EU AI Act per-system inventory, AI impact assessments, vendor-AI term review and a policy starter. A board briefing built on it covers:

  • An inventory of AI systems in use, with each one's role under the AI Act (provider, deployer) and risk category.
  • The literacy record for staff who use AI tools, since national authorities supervise Article 4 from 2 August 2026.
  • Contract terms on agent actions: authority, error, indemnity, audit rights, data use.
  • The incident process: who is told when an agent acts wrongly, and what the log shows.
  • Ownership: which executive owns AI policy, and how the board hears about it.

What corporate clients now expect

The commercial pressure is explicit. Axiom's survey of 528 in-house leaders (9 July 2026) found 92% expecting or negotiating AI-related rate cuts from outside counsel, while only 7% use and measure AI at scale themselves. Salesforce rolled Legora out to its global legal department on 16 September 2026, and UBS's legal team narrowed 80 candidate use cases to three with a 140% ROI target. The in-house view is set out in AI agents for in-house legal teams. For a corporate practice the response is to be able to show, per matter, which housekeeping items the agent drafted, who approved them and how long the review took. That record is also the answer when a client asks for the AI discount.

Setting this up in a corporate team

  • Choose the DMS connector and confirm the permission model with IT and the vendor in writing.
  • Write the housekeeping agent specification from the table above, with the gates unchanged, and run it on one entity group for a quarter.
  • Adopt the minutes prompt and the "from the pack, not from memory" rule; get recording consent into the board's standing procedure.
  • Fix the arithmetic rule: spreadsheet or tool computes, model drafts around the cell.
  • Prepare the board briefing pack on AI oversight using the inventory and impact-assessment templates, and log the team's own training for Article 4.

Corporate lawyers have spent careers building processes that produce the same document correctly every year. A housekeeping agent is that process with a faster clerk and a slower approval, and the approval is the part that matters.

Frequently asked questions

Can AI draft board minutes?

Yes, from the right inputs: the agenda, the board pack, the recording or the company secretary's notes, and the firm's minute template. The model produces a draft with each resolution and each recorded discussion tied to an agenda item and a source. It must not draft from memory of what boards typically resolve. Consent to any recording is obtained before the meeting, and the chair and the lawyer review every line.

What is a corporate-housekeeping agent?

A scheduled workflow that reads the entity register and the document management system through a permission-bound connector, checks upcoming filing and renewal deadlines, drafts the routine documents (annual resolutions, register updates, confirmation filings) from templates into a review queue, and reports what it did. It never files, sends or signs; a lawyer approves each item. Anthropic's Claude for Legal corporate plugin and firm-built n8n workflows both follow this pattern.

Is Claude Opus 5 better for corporate work?

Anthropic says so: its July 2026 release notes claim a clear step up on legal agent work with the biggest gains in corporate governance and arbitration, first-turn NDA redlines nearly double Opus 4.8, and financial-modelling accuracy up nine points. Those are vendor claims. The practical answer is to test it on your own board pack and cap table against whatever you use now, with a lawyer checking every number.

What should a board ask its lawyers about AI oversight?

Which AI systems the company deploys and where they sit under the EU AI Act, whether the Article 4 literacy duty and the Article 50 transparency duties that applied from August 2026 are met, what contracts say about agent actions and liability, what the incident and audit process is, and who owns AI policy. A per-system inventory and an AI impact assessment template are the first two documents to produce.

Put this into practice

AI Lab for Lawyers: a live, hands-on course for practising lawyers

4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.

Reserve a seat · €200 Maven money-back guarantee · certificate of completion