Superintelligent Agents for Dog Boarding: Reservations and Vaccine Files

A dog boarding kennel can use an AI agent today to handle the repetitive parts of reservations: answering availability questions, collecting stay details, sending confirmations and reminders, and flagging bookings that need a person. Vaccine record checks can also be partly automated, by reading what an owner uploads, comparing it against the kennel's requirements and listing what is missing or expiring. The agent should not decide whether a dog is fit to board or whether a record is valid until its source has been verified, because a wrong health decision costs far more than a slow reply.

Key takeaways

• Reservations are the safest first task: availability, stay details, confirmations and reminders follow clear rules that a staff member can audit.
• Vaccine checks work best as a comparison step, where the agent lists what is present, missing or near expiry, and a person confirms the outcome.
• An uploaded photo or PDF is not a verified record. Store the source, the date it was received and who approved it before any agent relies on it.
• Approval boundaries should be written down before the agent goes live, including what it may say to owners and what it may never decide.
• A good first engagement names one role, one set of records and one success measure that a manager can check in the kennel's own system.

Where boarding kennels lose time

Most of the friction in a boarding business happens between the first message and the drop-off. An owner asks whether there is room over a holiday weekend. Someone checks the calendar, asks for the dog's name, breed, feeding routine and medication, then asks for vaccine paperwork. The owner sends a photo of a card taken at an angle, or a link to a clinic portal. A staff member has to read it, decide whether it covers what the kennel requires, write back, and then remember to chase the missing item a few days before arrival.

Each of these steps is simple. Together, they fill the front desk and create gaps: a request that sat unanswered over a weekend, a record that was accepted without anyone noticing the expiry, or a medication note that lived only in a text thread. An AI agent is useful here because it can hold the context of a booking across many messages, follow the same checklist every time, and keep the open items visible. It is not useful when it is given access to everything and asked to improvise.

What an agent can handle in reservations today

Reservation handling has a clear shape, which makes it a good starting point. The agent can answer availability questions from the kennel's own calendar, offer the next open dates, collect the details the kennel needs, and send a confirmation that repeats the key terms. It can also send reminders before drop-off and after pickup, and route anything unusual to a staff member.

A practical scope for the first version looks like this:

• Answer availability questions using the live calendar, never from memory or a cached spreadsheet.
• Collect a fixed set of stay details: owner contact, dog name, dates, feeding and medication notes, and emergency contact.
• Create a draft booking that staff approve, rather than a confirmed booking the agent sends on its own.
• Send confirmation and reminder messages from approved templates, with the kennel's policies quoted from the current version.
• Flag requests for special handling, such as dogs that are reactive with other dogs, senior dogs with mobility limits, or late-night pickups, for a person to review.
• Log every change with a timestamp and the channel it came from, so a manager can reconstruct the booking later.

The agent should not promise space it cannot see, waive a deposit, change a cancellation rule, or accept a medication change without a staff member reviewing it. Those decisions belong to the people accountable for the animal.

Vaccine record checks: a step-by-step plan

Vaccine records are the part of boarding where automation can save the most time, and where careless automation does the most harm. The useful pattern is to let the agent do the comparison and let a person make the decision. Requirements differ by facility, by local rules and by the vaccines a clinic administers, so the kennel should write its own list before anything is automated.

A workable sequence for the check:

• Write the kennel's current requirement list, naming each vaccine, the accepted format and what counts as current for the stay dates.
• Ask the owner to upload the record from the veterinary clinic, or to confirm the clinic that holds it, so the source is known.
• Read the record and extract only the fields the list needs: vaccine name, administration date, expiry or due date, and the issuing clinic.
• Compare each extracted field against the requirement list for the booked dates, and mark each item as present and covering the stay, present but expiring during the stay, missing or unreadable.
• Send the owner a plain message that names exactly what is missing or unclear, and what format would resolve it.
• Hand the completed comparison to a staff member, who accepts, rejects or requests more information. Only the staff member changes the booking status to cleared.
• Store the source file, the comparison and the decision together, so a later audit can see what was known at the time.

Notice that the agent never marks a dog as cleared. It prepares the evidence and the next message. That separation is the main design choice in this workflow.

Why vaccine files need verified sources first

A vaccine record is a health document, and a health document is only as reliable as its origin. A photo of a card can be cropped, outdated, belong to a different dog, or show a vaccine that was given but not recorded in the way the kennel requires. A clinic portal may show a status that does not match the paper in the owner's hand. An owner may type a date from memory. None of these problems are rare enough to ignore.

This is where a stronger agent creates risk. If an agent is given broad access to the kennel's records, the owner's messages and the booking system, it can be pushed by ordinary text or by a deliberate attack to treat a message as instruction. The OWASP guidance on prompt injection describes how untrusted input can alter what a model does when that input is mixed with its instructions, and the same reasoning applies to a document an owner uploads. Text inside a file is data to be checked, not a command to be followed.

For that reason, the kennel should separate three things. The first is the verified source: the clinic record, confirmed by the clinic or by a document the clinic issued. The second is the owner's claim, which is useful but unverified. The third is the agent's comparison, which is a draft. Booking decisions should rest on the first and be made by a person who can see the difference between the three.

The NIST AI Risk Management Framework is a useful reference for this kind of planning. It asks organizations to map the context a system operates in, measure its behavior against defined criteria, and manage the risks that remain. For a kennel, that translates into a short written list of what the agent may read, what it may propose, and what must be confirmed by a person before it counts.

Where approval boundaries belong

Authorization is the second layer. An agent should only see the records and actions its role needs. The reservation agent needs the calendar and the booking draft queue. It does not need the full billing history, the staff schedule or the medical notes for dogs not in the current request. The OWASP authorization guidance describes the general principle: check access on every request, deny by default, and keep permissions narrow enough that a mistake has a small blast radius.

In practice, a kennel should decide in advance:

• Which fields the agent can read and which it can never see, such as payment details.
• Which actions it can take on its own, such as sending an approved reminder.
• Which actions require a staff approval, such as confirming a booking or accepting a record.
• Which requests must go straight to a person, such as a medical question, a dog that bit another dog, or an owner disputing a fee.
• Who can change these rules, and how changes are recorded.

These boundaries should be reviewed on a schedule, not only when something goes wrong. Staff change, clinics change their portals, and the kennel's own policies evolve.

An illustrative week at the front desk

The following scenario is illustrative and does not describe a real kennel. A small boarding business receives a holiday weekend request from an owner who has a senior dog on a daily medication. The agent checks the calendar, offers two open dates, and collects the dog's details. Because the request includes a medication, it creates a draft booking and flags the note for a staff member. The owner uploads a vaccine record taken on a phone. The agent reads the vaccine names and dates, finds one vaccine due to expire during the stay, and writes to the owner asking for a current copy or a confirmation from the clinic. It does not mark the dog as cleared.

The staff member reviews the draft in the morning, calls the clinic to confirm the expiry date, and approves the booking once the record is verified. The reminder sequence then runs from the approved template. At pickup, the agent logs the return time and prompts a short check on the medication schedule that a person reviews. Nothing in this week required the agent to make a health judgment. It required the agent to keep track, to compare and to ask clearly.

How Be Human approaches the first engagement

Be Human designs and operates this kind of personalized AI infrastructure as a managed service. A first engagement defines one useful role for the kennel, usually the reservation desk or the vaccine check, along with the memory and tools that role needs, the approval boundaries staff must hold, and an observable success criterion, such as the share of bookings that reach a complete record without a manual follow-up. Managed operations can then include monitoring of how the agent handles real requests, recurring evaluations against the kennel's own requirement list, permission reviews when staff or clinic integrations change, and controlled improvements. The work is remote, in English or Portuguese, and the kennel keeps control of its identity, memory and channels. If you want to find the first task worth handing to an agent in your own business, take the free Be Human AI diagnostic.

Frequently asked questions

Can an AI agent confirm a dog's vaccines for boarding? It can read the records the owner provides, compare them against the kennel's written requirements and list what is missing, expiring or unclear. The confirmation itself should be made by a staff member who has verified the source, usually the clinic. Treating the agent's comparison as a draft keeps the health decision where it belongs.

What should a kennel automate first? Reservation handling is usually the safest starting point because availability, stay details, confirmations and reminders follow clear rules. Vaccine comparison can come next, once the requirement list is written and the verification steps are in place. Starting with a single role makes it easier to measure whether the agent is helping.

What happens when an owner sends a record that looks wrong? The agent should not try to correct it or guess at the missing information. It should send a plain message naming the specific problem and the format that would resolve it, then route the case to a staff member with the source file attached. The file itself is treated as data, so any instructions inside it are ignored.

Does the agent need access to all of the kennel's data? No. The reservation role needs the calendar, the booking draft queue and the fields required for a stay. Payment details, staff schedules and unrelated dog records should be outside its permissions. Narrow access limits the damage a mistake or a manipulated message can cause, and it makes permission reviews much easier to carry out.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.