Superintelligence Prep for Animal Shelters: Adoption Applications
Adoption applications, reference checks and foster scheduling for animal shelters, with a clear line for where staff keep the final decision.

For an animal shelter, the practical form of superintelligence is not a system that decides which animal goes to which home. It is an agent with a written memory of shelter rules that receives adoption applications, checks them for completeness, requests references and proposes foster visit slots, while named staff keep the final adoption decision and the reasons behind it. That separation is what makes the work safe to hand over.
Key takeaways
- Start with intake completeness, reference requests and visit scheduling, not with the adoption decision itself.
- Give the agent read access to the application and a written checklist, but no authority to approve, deny or promise a placement.
- Schedule foster visits against the capacity rules the shelter already uses, with a staff member confirming each slot.
- Log every outreach, reply and missing item so the coordinator can audit what the agent did and why.
- Review permissions and the checklist on a fixed schedule, because program rules, volunteers and staffing change.
What an adoption application agent should handle
An adoption application agent works on the paperwork around a decision, not the decision itself. Most of the coordinator's week goes to receiving forms, chasing missing fields, emailing references, finding a time for a meet-and-greet and writing summaries that someone else has to read. Those tasks are repetitive, easy to check and easy to reverse, which makes them a sensible first place to use an agent.
Tasks the agent can usually own, once the shelter has approved the rules, include:
- Acknowledge each application and tell the applicant what happens next, in the language they used.
- Check the form against the published requirements for each program, such as dogs, cats, rabbits or foster-to-adopt.
- Request missing documents and send reference requests from approved templates, then follow up on the schedule the coordinator set.
- Propose meet-and-greet or foster visit slots from the calendars staff already maintain.
- Write a one-page summary that separates verified facts from anything the agent merely observed.
The agent should not approve, deny or rank applicants for a specific animal, judge whether a household suits an animal's history, waive a fee or make promises about timing that depend on someone else's decision. Those belong to people. The same split between intake and decision appears in other intake-heavy work, such as behavioral health clinic intake, where the scheduling layer can move quickly while clinical judgment stays with licensed staff.
Writing the intake checklist before any automation
Most failed automations start with a checklist that exists only in one coordinator's head. Before connecting a form inbox or a calendar, write down the rules the shelter already applies. An agent that receives a vague instruction will fill the gaps itself, and the gaps are where errors appear.
A workable order for building the checklist:
- List every field the shelter needs and mark which ones are mandatory for each program or animal type.
- Write the reason behind each requirement, so the agent can explain it to an applicant without inventing policy.
- Mark the answers that trigger a human review, such as a rental agreement that restricts animals, a prior surrender on record, or a note about aggression in an animal's current file.
- Define what complete means, and what happens to an incomplete file after a period the coordinator sets.
- Name the person who owns each exception, and the channel they use to answer.
- Keep the checklist in one versioned document, and record the date of each change.
Once the checklist is written, test it against real past applications, including the messy ones. If two experienced staff members would classify the same file differently, the rule is not ready for an agent. Fix the rule first, then automate it.
Reference checks without collecting more than you need
Reference checks are useful, but they are also where a shelter holds personal data about people who never asked to be in a database. Ask only for what changes the decision: the reference's name and contact, the specific question the shelter needs answered, and the consent wording that covers the request. Organizations that process personal data in Brazil should review their obligations under the LGPD, and the national data protection authority publishes guidance on them. Similar rules apply in many other places, so confirm the requirements with counsel before the agent sends its first message.
Practical rules for the reference step:
- Send a reference request only after the applicant has consented to that specific request.
- Use a fixed set of reference questions, so answers can be compared across applications without rewording.
- Store each reply in the applicant file with its date and the sender's role, such as landlord or veterinary clinic.
- Do not let the agent summarize a reference in a way that adds judgment the reference did not give.
- Restrict or delete an applicant's data when they withdraw, according to the shelter's retention policy.
Expiring documents deserve their own tracking, since a veterinary record or a rental letter can go stale while the file sits open. The same pattern appears in credentialing work, where the value is knowing what expires and who must act, not deciding whether the document is acceptable.
Foster scheduling: capacity first, calendar second
Foster and visit scheduling looks simple until a volunteer cancels, a foster home closes for a week or an animal needs a quiet room that is already booked. The agent should check capacity rules before it checks the calendar. Those rules include how many visits a foster home can host, which animals can be shown in which rooms, which volunteers are trained for which species, and when a staff member is available for a health-related handoff.
Illustrative scenario: a small shelter has two foster homes certified for cats only and one volunteer who handles dog visits on weekends. An applicant asks for a weekend meet-and-greet with a dog. The agent checks the dog's visit rules, finds one open volunteer slot, and sends the coordinator a note: "Saturday morning slot proposed for the applicant, who has confirmed availability. Please confirm or reassign." The coordinator confirms, the agent records the confirmation in the file, and the applicant receives the time. The agent never offers a placement or implies that the visit is an approval.
Foster programs also need a clear rule for cancellations and no-shows. Decide in advance who may rebook, how many attempts the agent may make, and when the case goes back to a person. A rebooking that the coordinator has not approved is a common source of frustrated applicants and confused volunteers.
Where staff keep the final adoption decision
The boundary between agent and staff should be written down before go-live, and the people who will live with it should agree to it. The following decision guide is a useful starting point, to be adapted to the shelter's own policies and to local law.
- Recording facts that come with a source, such as a completed form, a dated reference reply or a received document: the agent may record them and flag gaps.
- Drafting summaries or comparisons between applicants for the same animal: the agent may draft, and a coordinator reviews before anyone else sees the output.
- Booking a visit into a slot the coordinator has pre-approved: the agent may book and must log the booking.
- Approving, denying, ranking or choosing an applicant for a specific animal: a named staff member decides and records the reason in the file.
- Any message touching medical, behavioral or safety concerns: route it to a human immediately and do not let the agent answer the substance.
The same discipline shows up in family law intake, where the agent can collect and organize dates and documents while a professional decides what they mean.
Be Human's managed approach starts with one role, usually intake and scheduling, and defines its memory (the shelter's checklist, program rules and approved templates), its tools (the form inbox, the calendar and the file store), its permissions and the approval points above. Be Human operates the agent remotely in English or Portuguese, reviews its logs, updates permissions when staffing changes, and handles incidents such as a misrouted message. The first success criterion is written before launch, for example that every open file has a checklist status and every decision has a named person attached.
Protecting permissions, logs and the messages applicants send
Applicants write long free-text answers, and some of that text may try to change how the agent behaves, whether by accident or on purpose. The safe rule is to treat every applicant message as data, never as instructions. The OWASP prompt injection guidance describes this class of risk and the defensive patterns that reduce it. Keep the agent's own tools narrow as well, so that an injected request cannot do more than the workflow allows.
Authorization should follow least privilege. The intake agent can read the form, write notes and propose slots, but it cannot change an animal's status record, send a message that states a decision or export the applicant database. The OWASP authorization guidance is a useful checklist for reviewing which identities can perform which actions, and the NIST AI Risk Management Framework helps document who oversees the system and how failures are handled.
Logs are what make the system governable. For each action, keep a record of the template used, the data read, the time, the person who approved the step and any exception raised. Without these records, a complaint from an applicant cannot be investigated, and the checklist cannot be improved from real cases. Test the agent before launch with incomplete forms, contradictory answers and a free-text message that instructs the agent to approve the application. The pass condition is that the file reaches a person.
Where a shelter should start
Start with one role, one animal program and one outcome that the coordinator can check, such as every new application receiving a complete checklist status within a window the shelter sets. Add reference requests and foster scheduling only after the logs show the agent following the checklist and escalating the cases it should escalate. Keep facts with the agent and decisions with people, and widen the agent's scope only when a review shows that the boundary is holding. Many shelters find that the first useful task is smaller than they expected, and that is a good sign. If you want to find the first task worth handing to an agent in your shelter, take the free Be Human AI diagnostic shown below.
Frequently asked questions
Can an AI agent approve an adoption application? No, and it should not be set up to try. In a sound design, the agent checks completeness, requests references, summarizes what the file contains and proposes visit times. A named staff member makes the decision and records the reason, so the shelter can explain it later.
Does the agent contact references on its own? It can, but only after the applicant has consented to that specific request and only with the reference questions the shelter has approved. Each reply should be stored in the applicant file with its date and the sender's role. The agent should not turn a reference's answer into a judgment that the reference never gave.
How does foster scheduling avoid double bookings and overloaded volunteers? The agent checks capacity rules before the calendar, including which homes are certified for which animals and how many visits each volunteer can host. It proposes only pre-approved slots, and a coordinator confirms each booking. Every change is logged, so a conflict can be traced to its source.
What happens when an applicant's answer raises a medical, behavioral or safety concern? The agent should route the message to a human immediately and avoid replying to the substance of the concern. The message and the routing step are logged, and the coordinator decides the next step. This rule belongs in the written checklist before the system goes live, not in a prompt written after the first incident.
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.


