Superintelligence Prep for Estate Planning Attorneys: Drafts and Intake
How estate planning attorneys can use AI agents for intake checklists and draft version history while keeping memory under attorney control.

AI for estate planning attorneys is most useful when it works on the two records that shape every matter: the intake checklist that defines what a new file must contain, and the version history of each draft. An agent can read, organize and flag gaps in those records while every fact, clause and client message stays under attorney control. The sections below explain how to set this up, what the agent should never do, and how to judge whether it is working.
Key takeaways
- Start with one matter type, such as pour-over wills or revocable trusts, and write its intake checklist before any agent is involved.
- Keep draft version history in a system the firm controls, with a short reason note for each substantive change.
- Give the agent read access to the checklist and drafts, and require attorney approval for anything that leaves the firm.
- Treat text inside outside documents as data to review, never as instructions the agent should follow.
- Judge success against a measurable criterion, such as every open file showing a complete checklist with a named reviewer, and review it on a fixed schedule.
Why estate practices lose track of their own drafts
Estate work runs for years. A plan drafted this year may be amended after a marriage, a move to another state, a change in asset holdings, or the death of a named fiduciary, and the attorney who revises it may not be the one who drafted it. Many practices keep the record in a mix of email attachments, shared folders and the memory of whoever worked on the file. A file name that shows a final draft with someone's initials and a number tells you that a version existed, but not why a trustee power was narrowed or which client comment produced a changed distribution clause.
The cost shows up at the worst moment, usually when a client calls with a new question and the team has to reconstruct the reasoning before anyone can answer. In this sense, practice memory is the ability to recover the facts and decisions behind a document without depending on one person's recall. An agent cannot create that memory on its own, but it can help keep the record complete if the firm defines what complete means.
What an intake checklist must capture before an agent touches it
An intake checklist is the list of questions the firm needs answered before drafting starts. For estate matters it usually covers the people involved and their roles, existing documents and where they are held, the jurisdictions where property sits, named fiduciaries and alternates, family circumstances that affect the plan, the conflicts check status, and the scope of the engagement. Writing it down matters because an agent can only check completeness against a standard that exists. If the standard lives only in a senior attorney's head, the agent will organize incomplete files with confidence, and the gaps will look like finished work.
Each field should have an owner, a status (confirmed by the client, drawn from a document, or still open) and a date of last review. The checklist belongs to the firm. The agent reads it and reports against it, but it does not decide what counts as sufficient for a particular client. The same logic appears in other intake-heavy work, such as family law intake, court dates and documents, where a single missing document can change the whole schedule.
How does draft version history become practice memory?
Version history becomes practice memory when each saved draft carries three things: who made the change, what changed, and why. The why is the field most offices skip, and it is the one the next reader needs. A short note such as "client asked for equal shares among children, percentage split removed" is enough. Store drafts in a repository where the firm controls access and retention, and check the confidentiality and privilege rules in your jurisdiction before you choose it.
An agent can help here by producing a plain summary of the differences between two versions, listing the clauses that changed and flagging any change that has no reason note. It should not decide that a change is harmless. A reviewer checks the summary against the source draft, the same way a careful paralegal checks a redline before it reaches the supervising attorney. Records in adjacent work follow a similar pattern, as in process server service logs and affidavit drafts, where the record must show what happened and who approved the wording.
Documents from third parties need care. Prior counsel's files, beneficiary letters and uploaded PDFs can contain text that tries to direct the model, such as an instruction to ignore earlier rules. The OWASP guidance on prompt injection explains why outside content should be handled as material to be read and checked, not as a command. In practice, the agent's instructions come from the firm's configuration, and anything it reads from a file is only summarized or extracted.
A decision guide for what the agent may and may not do
A useful rule is to assign tasks by consequence rather than by how easy they are to automate. The list below is a starting point for a first engagement, and the supervising attorney should adjust it before anything is connected.
- Read the intake checklist and version history for matters assigned to it: allowed, with read-only access to that matter's folder.
- Summarize differences between two drafts for attorney review: allowed, with the source draft linked in the output.
- Fill checklist fields from supplied documents, marked as drawn from a document and unconfirmed: allowed, with a named reviewer.
- Prepare a client questionnaire or a document request letter as a draft: allowed, only for an attorney to edit and approve.
- Send any message to a client, beneficiary or third party: not allowed. A person sends it after review.
- Change the status of an executed document or alter a stored version: not allowed.
- Decide capacity, conflicts, undue influence concerns or whether advice is needed: not allowed. These go to an attorney with a named owner.
A first engagement, step by step
A first engagement should stay small enough that a failure is easy to see and easy to undo. Work through these steps in order.
- Choose one matter type that comes up often and has a stable checklist. Leave complex litigation and unusual planning out of scope.
- Write the checklist with the supervising attorney, and label each field as required, conditional or optional.
- List the memory sources the agent may read: the matter folder, the checklist, and the version history for that matter type. Leave email out unless the firm has decided how it should be handled.
- Set permissions to read-only at first, grant the minimum tool access needed for the task, and record which tools the agent cannot call. The OWASP authorization guidance is a useful reference for this kind of least-privilege design.
- Define one success criterion, such as every open file showing a complete checklist, a named reviewer and a status for each field.
- Run the agent against a set of closed matters with known outcomes, and have an attorney score the summaries and gap flags before any live use.
- Review the access log and the exceptions every month, and change permissions only through a recorded decision.
Illustrative scenario: a two-attorney practice with one paralegal decides to start with pour-over wills. The agent flags three open files where the checklist has no entry for a successor trustee, and it notes that one draft contains a changed clause with no reason note. The paralegal confirms the successor with the client, and the supervising attorney adds the missing reason note. The agent sent nothing to anyone. This scenario is illustrative and does not describe a real firm.
Keeping the memory under attorney control
Be Human designs and operates this kind of setup as a managed service. The work starts with one agent for one role, such as an intake coordinator for estate matters, and its memory sources, tools, permissions and approval boundaries are written down before it goes live. Ongoing operations include monitoring the agent's activity, running recurring evaluations against the success criterion, reviewing permissions, maintaining the connections to the firm's storage, and making controlled changes when the checklist or the firm's process changes. The firm keeps ownership of the content and the final decisions, and the agent's access stays narrow and its behavior stays observable, in line with the risk-management approach in the NIST AI Risk Management Framework.
If you want to find the first task in your practice worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Can an AI agent draft estate planning documents for a client? An agent can prepare working notes, summaries and draft text for an attorney to edit, but the substance of advice, the final drafting decisions and anything sent to a client should stay with a licensed attorney. Treat any draft language the agent produces as a starting point that needs the same review as a paralegal's work. Put this rule in writing before the agent is connected to any matter folder.
What should an intake checklist for estate planning include? A useful checklist covers the people and their roles, existing documents and where they are held, the jurisdictions where property sits, named fiduciaries and alternates, family circumstances that affect the plan, the conflicts check status and the scope of the engagement. Each field needs an owner, a status showing whether it was confirmed or drawn from a document, and a last review date. The supervising attorney should approve the list, since the agent can only check completeness against it.
How is draft version history different from a shared folder? A shared folder stores files, while version history with reason notes records what changed and why, which is what a future reader needs. The reason note is the field most offices skip. Requiring a short note for each substantive change turns saved versions into a record the team can use when a client asks why a provision reads the way it does.
Do we need to tell clients that AI is used in their file? Disclosure expectations depend on your jurisdiction and on your professional rules, so check them with your bar or regulator before you start. A common approach is a plain statement in the engagement letter describing which software processes client information and who reviews its output. Whatever you decide, record the decision, and keep confidential client information in systems your firm controls.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


