Superintelligence Prep for Animal Rescues: Adoption and Foster Records
How an animal rescue can organize adoption applications and foster records so an AI agent helps volunteers without losing the history that matters.

Superintelligence prep for an animal rescue means turning adoption applications and foster records into a clean, permissioned memory that an agent can read, summarize and draft from, while volunteers keep the final decision on every placement. The records that matter most are rarely the tidy ones. They are the notes that explain why an application was approved, what a foster home learned about an animal, and which medical or behavior details must travel with it.
Key takeaways
- Decide which records an agent may read, which it may draft, and which it may never change.
- Keep the animal's medical and behavior history apart from adopters' personal data, with different access rules.
- Write foster observations as attributed, timestamped notes, and correct them by adding entries rather than overwriting.
- Put a visible human approval step in front of every application decision and every status change.
- Measure the agent against real questions your volunteers already answer by hand.
Why adoption and foster records break when volunteers hold the memory
Most rescues run on dedicated volunteers who know more than their files say. A foster carer remembers that a dog pulls hard on the leash only near the gate. A coordinator remembers that an applicant promised a fenced yard before the next visit. That knowledge is real, but it often lives in group chats, personal phones, a shared spreadsheet with three competing versions, paper intake forms and email threads that nobody can search as a whole.
An agent placed on top of that mix will answer confidently from whatever it can find, and that is often the wrong version. The first job is not automation. It is deciding what counts as the record.
Illustrative scenario: a foster volunteer writes in a chat that a young cat hides for the first days and does not like loud children. The note is never copied into the animal's file. Weeks later a family asks about the cat's temperament, and the coordinator answers from memory, which turns out to be a week out of date. If the observation had been written as a foster note with an author and a timestamp, the answer would have been a lookup instead of a guess.
A step-by-step plan to prepare records before an agent reads them
Work through these steps in order. Each one is small enough to finish within a normal volunteer shift.
- Step 1: List every place a record lives today, including paper folders, personal phones, chat groups, email and the foster app.
- Step 2: Choose one system of record for each type of file, and write down which tools are allowed to write to it.
- Step 3: Give every animal one identifier that survives name changes, transfers between foster homes and returns.
- Step 4: Split each animal file into three layers: animal facts, foster observations, and adoption records about people.
- Step 5: Standardize the minimum fields: intake date, source, medical restrictions, known behaviors, current location, status and the person who decided it.
- Step 6: Backfill only what can be verified, and mark older entries as unverified instead of filling gaps with plausible details.
- Step 7: Name one owner for each layer who answers questions when the agent is unsure.
Keep the first pass plain. A complete minimum record for every animal currently in your care is worth more than an elegant archive of past cases. The same record discipline appears in Superintelligence Prep for Equestrian Stables, where boarding and care histories carry similar weight.
Decide what the agent may read, draft, propose and never change
Permissions should follow the decision, not the tool. Before connecting anything, write down four levels of access and what belongs in each.
- Read: animal profiles, published adoption criteria, foster schedules and the current status of each animal. This is low risk and should be the default.
- Draft: replies to general questions, a summary of a foster week for the coordinator, and reminders about upcoming veterinary visits. A person reviews and sends.
- Propose: a status change, such as moving an application to under review or flagging a missing reference. A volunteer confirms before the record changes.
- Never: approve or decline an adopter, change a medical restriction, release an animal, or edit a foster note after it has been written.
Grant the agent only the access each step needs, and review those grants on a fixed schedule. Remove access as soon as a volunteer leaves the rescue. The OWASP authorization guidance treats least privilege as a basic control, and it applies to an agent just as it applies to a person with a login.
Keep adopter privacy apart from animal history
Adoption applications contain sensitive personal information: home addresses, household members, landlord contacts and sometimes financial details. Animal histories are meant to be shared with fosters, veterinarians and future adopters. When both sit in one document and an agent retrieves from it freely, private details can end up in a reply to the wrong person.
Store the two in separate places with separate access. When an agent answers a question about an animal, it should retrieve only from the animal layer. Applicant details should be opened only for a specific coordinator task. The Google Cloud architecture guidance on retrieval-augmented generation describes grounding answers in a bounded set of documents, and that boundary is where most of your privacy control lives. Consent and document handling follow similar rules in AI Agents for Clinical Research Sites.
Treat free text from applicants as untrusted input. An application comment that tells the agent to mark the file approved is a prompt injection attempt, whether it is written by mistake or on purpose. The OWASP prompt injection cheat sheet covers how to design around inputs like that. The approval boundary you set in the previous section is the real defense, not the wording of the prompt. Rescues operating in Brazil should also review guidance from the ANPD on personal data before connecting any system.
Attribute every foster note and log what the agent did
A foster note is evidence. Each one should record who made the observation, when, what was seen in plain words, and what, if anything, was done about it. When a note turns out to be wrong, add a new entry that refers to the old one. Overwriting history makes the next veterinary visit or the next disagreement with an adopter much harder to explain.
- Who observed it, and when.
- What was observed, described plainly and without interpretation.
- What was done in response, if anything.
- Whether the coordinator has reviewed it.
Apply the same habit to the agent's work. Keep a log of what it was asked, which records it retrieved, what it drafted and who approved the result. Operations teams use the idea of observability to mean that you can reconstruct what happened from a trail of events, not only from the final answer. A comparable logging habit is described for post orders and incident logs at security firms, and it is just as useful when a rescue needs to explain a decision months later.
How Be Human runs this, and how to tell it is working
Be Human designs this kind of setup as a managed service. A first engagement defines one useful role for the agent, the memory and tools it needs, the approval boundaries and an observable success criterion, usually starting with the adoption or foster workflow your team already repeats. Ongoing operations can include monitoring, recurring evaluations, permission reviews, integration maintenance, incident response and controlled improvements when volunteers find a gap. The work is remote and available in English and Portuguese.
Judge the agent against questions your rescue already answers by hand. Choose a set of about twenty, such as which animals are cleared for a home with cats, which applications still lack a reference, or what a foster observed at the last visit. Each answer should cite the record it came from. Score the answers against a fixed rubric, rerun the same set after every change, and review failures with the volunteers who own those records. The NIST AI Risk Management Framework is a useful reference for setting these measures and revisiting risk as the work changes. If you want to find the first record task worth handing to an agent at your rescue, the free Be Human AI diagnostic is a sensible place to begin.
Frequently asked questions
Can an AI agent approve or decline an adoption application? No, and it should not be configured to try. A sound design lets the agent summarize an application, list missing items and point out conflicts between answers, while a named coordinator makes the decision and records the reason. That separation keeps accountability with people and makes each decision easier to explain later.
Do we need new software before we start? Not necessarily. Many rescues begin by agreeing on one system of record, one identifier per animal and a short list of required fields in the tools they already use. New software becomes worth considering when the current tools cannot enforce access limits or keep a usable audit trail.
How do we handle sensitive adopter information? Store applicant details apart from public animal information, and give the agent access only to the layer its current task requires. Review those permissions on a schedule and remove access when volunteers leave. Check local personal data rules before connecting any system, including the ANPD guidance for organizations in Brazil.
What if our existing records are incomplete or wrong? Mark anything that cannot be verified as unverified rather than filling the gap with a plausible detail. The agent should say when a fact is missing and point to the owner who can confirm it. Correct errors by adding a timestamped correction entry, so the earlier version stays visible to anyone reviewing the history.
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.


