Personalized AI infrastructure
AI for Cooperatives: Keep Member Records in Memory You Own
Keep member records and shared decisions in memory your cooperative owns, so an AI agent serves members without changing bylaws, roles or approval rules.

A cooperative can let a stronger AI agent serve members while keeping member records and shared decisions in memory the cooperative owns. The agent reads and writes that memory through narrow, logged tools, and every answer points back to a record or a resolution. The bylaws, the roles and the approval rules stay exactly where the members put them.
Key takeaways
- Member records, shared decisions and governance rules should live in a memory the cooperative owns, not inside a vendor's chat history.
- The agent reads and writes through named tools with scoped permissions, so every change has an owner and a log entry.
- Assembly resolutions and board decisions are stored with their source, approving body, effective date and replacement status, so the agent never reinterprets them.
- Start with one member-facing task, such as answering membership status questions, before any task that moves money or changes a member's standing.
- Review permissions and records on a fixed calendar, because membership rules, board members and staff change over time.
Why owned memory matters more for a cooperative
A cooperative is governed by the people who joined it. Members hold membership, make contributions, vote in the general assembly and expect the board to apply the statutes consistently. Those facts usually live in several places at once: a membership form, a spreadsheet of contributions, email threads about exceptions, and minutes stored in a shared folder that only two people know how to search.
A more capable agent does not fix that scatter. If it answers from whatever it can find, it may quote an old policy, treat a draft as a resolution, or present a member's status with more confidence than the records justify. Owned memory means the cooperative decides what the agent may hold, who may read each part, how long records are kept, and who corrects them when they are wrong.
Governance rules and memory are different things. The bylaws and assembly resolutions define what the cooperative has decided. Memory holds the facts and decisions in a form the agent can retrieve. The agent works from the rules; it does not write them. For a fuller treatment of how a board draws that boundary, see Superintelligence Governance: A Committee Charter for Small Companies.
Three kinds of memory to keep apart
Most cooperatives mix three kinds of information. Keeping them in separate stores, or at least in separate permission scopes, prevents a routine member question from exposing a board discussion.
- Member register: identity, membership status, contribution history, contact preferences and recorded consents.
- Decision record: assembly resolutions, board minutes, approved policies and exceptions, each with the body that approved it, the date it took effect and whether a later decision has replaced it.
- Operating context: current procedures, staff-written answers to common questions, forms, templates and the list of open requests.
The separation matters because access differs. A member may ask whether a contribution was recorded, but should not see another member's history or the minutes of a closed session. A staff member preparing an assembly may need the decision record in full. The agent should inherit the same split the cooperative already uses for its paper files, even if that split was informal until now.
Decide what the agent may read, write and only propose
Each record type needs one of four permissions. Write them down before any tool is built, because the tool design follows from this list.
- Read: public policies, approved member FAQs, and the member's own record once identity has been verified.
- Write, with a log: contact updates, new support requests, and preparation notes for staff. Each write names the caller and the source it relied on.
- Propose only: admissions, exclusions, changes to a contribution schedule, and any payment or distribution. The agent drafts the case with the relevant documents attached and routes it to the body the bylaws name.
- Never: reinterpret the statutes, override a resolution, or edit a historical minute.
The propose-only line needs the most care. An agent that drafts an exclusion letter with the relevant resolution attached saves time. An agent that sends that letter is making a governance decision. Thresholds can rise as evidence builds, as explained in Approval Thresholds for Superintelligent Agents: Raise Them Slowly, but the first version should default to a person deciding. Permissions should also be reviewed on a schedule, as set out in How to Run a Permission Review for Your AI Agents, Step by Step.
A step-by-step plan for the first records project
Start with one member-facing task and only the records that support it. The plan below takes a small cooperative from scattered files to a supervised first answer.
- Step 1: List every place where member facts or decisions live today, including personal inboxes, paper forms and spreadsheets kept by one person.
- Step 2: Name a records owner for each register and each decision type. An owner is a person, not a shared mailbox.
- Step 3: For each governance question members ask, write down which document controls the answer and who may interpret it.
- Step 4: Remove duplicate member entries and mark each record with the date it was last verified against a source document.
- Step 5: Define one tool per function, such as looking up membership status or listing effective resolutions. Each tool returns only the fields its role needs.
- Step 6: Write the success criterion in plain terms: the agent answers a status question with the record it used, and a staff member confirms the answer matches the source.
- Step 7: Run a supervised period in which staff review each answer before it reaches a member, and log every correction as a formal change request.
Illustrative scenario: two member questions in one week
Illustrative scenario, a fictional small dairy cooperative: a member asks whether her second contribution installment was recorded. The agent verifies her identity, reads her register entry, and replies with the recorded amount, the posting date and the source entry. It notes that the treasurer can correct any difference. The question closes without staff time.
Another member asks whether an exclusion decided at an earlier assembly can be reversed. The agent cannot judge that. It finds the resolution and the minutes that record it, confirms that no later resolution replaces it, and creates a handoff for the board secretary with both documents attached. The member receives an honest reply: the question has gone to the people the statutes name, and the next step is a written response from the secretary. Both interactions leave a log entry that staff can read later.
How shared decisions stay true over time
A decision record is only useful if it states which version is in force. Every resolution or policy should carry an identifier, the body that approved it, the effective date and a field showing whether a later decision replaced it. The agent should answer with the current version and name it, rather than the most recent document it happened to retrieve.
Logging is the other half. Each tool call should leave a trace: when it ran, which caller asked, which record it touched and what it returned. Observability practice in software, summarized in OpenTelemetry's observability concepts, treats this kind of trace as the basis for finding and explaining problems. For a cooperative, the same trace answers the question an auditor or a member will eventually ask: who saw this record, and why was it shown?
Member messages also need careful handling. A message can contain text that reads like an instruction, such as a request to ignore the procedure or to reveal another member's record. The agent should treat member text as material to answer, never as a command that changes its permissions. OWASP's authorization guidance points to the same principle: access is enforced by the system's checks, not by the wording of a prompt.
Privacy, consent and how Be Human approaches it
Member records are personal data in most jurisdictions. A Brazilian cooperative works under the LGPD, supervised by the ANPD, and cooperatives with members in other countries bring their own requirements. The cooperative must decide the lawful basis for each use, the retention period for each record, and what members are told when an agent helps answer their questions. An agent does not settle those questions; it inherits the answers the cooperative gives.
Be Human designs, implements and operates personalized AI infrastructure for companies, where AI agents keep their identity, memory, channels, tools, permissions and operating context under human control. A first engagement defines one useful role, the memory and tools it needs, the approval boundaries and an observable success criterion. Managed operations can then cover monitoring, recurring evaluations, integration maintenance, permission review, incident response and controlled improvements, so the memory and permissions stay under the organization's control while the agent keeps working. The service is remote, in English and Portuguese.
To find the first task worth handing to an agent in your cooperative, take the free Be Human AI diagnostic.
Frequently asked questions
Can a cooperative let an AI agent answer member questions? Yes, for questions the records can answer, such as membership status, recorded contributions or the text of an approved policy. The agent should verify the member's identity first, cite the record it used, and hand off anything that requires a judgment reserved to the board or the assembly. Staff should review early answers until the success criterion holds consistently.
Does the agent change the cooperative's governance rules? No. The bylaws and resolutions remain the authority, and the agent's role is to retrieve and explain them. Any change to a rule goes through the body the statutes name, and the agent records the new version once it is approved, marking the earlier version as replaced.
Who owns the member data and the shared memory? The cooperative should own it, which means it decides what is stored, who may read it, where it is kept and how it is exported or deleted. That ownership should be written into the service agreement, so that a change of provider does not change who controls the records.
Where should a cooperative start? Start with one member-facing question that the existing records already answer, and write the success criterion before building anything. Clean the register for that question, name a records owner, and run supervised answers for a set period before widening the scope of the agent.
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.


