Any vendor selling you autonomous property management is asking you to delegate actions that carry legal and financial weight. Before you say yes, get written answers to twelve questions covering five areas: what actually gets executed, what's forbidden, what's reversible, what's recorded, and who's accountable when something goes wrong. A good product answers all twelve happily.

What we’ll cover

I've been a landlord for nine years and I build autonomous property management software for a living — which means I know exactly which questions I'd dread if my product were built badly. Here they are. Ask every one, in writing, of any vendor you evaluate. Including us.

1. "If I don't log in for two weeks, what gets fully completed — not drafted, completed?" This one question sorts the market. AI that drafts messages "for your review" is a writing assistant; you're still the property manager. Demand a task-level list: rent collected and escalated? Maintenance triaged and dispatched? Tenant questions answered? If the honest answer is "we'll have suggestions waiting," you're buying a copilot at autopilot prices.

2. "Show me a complete closed loop, end to end, in the live product." Not a demo video — the actual product completing an actual task from trigger to logged completion. Ask them to break it mid-loop: what happens when the vendor doesn't answer, the tenant replies with something unexpected, the payment fails? Grace under failure is where real products separate from demos.

3. "Which of my lease's actual terms does the system enforce — my late fee, my grace period, my notice periods — versus generic defaults?" Autonomy that enforces someone else's rules is worse than no autonomy. The system should read your lease and run your terms.

4. "What actions is your AI permanently barred from executing, no matter how I configure it? Send me the list in writing." The most important question on this page. There are only three answer types: a specific written list (good); "you can configure anything" (that's not flexibility — it's the vendor declining to take a position on safety, leaving every consequence configured onto you); or vagueness (worse). Our list, published permanently: evictions, applicant denials, security deposit deductions, fair-housing-sensitive actions.

5. "Is that floor enforced in code, or is it an instruction the AI is told to follow?" These are wildly different. An instruction ("the AI is trained not to…") can fail, drift, or be prompted around. A code-level gate cannot. If the vendor doesn't understand the difference, that's your answer.

6. "Can the AI deny a rental applicant, and if so, how do you handle fair housing?" If any part of the accept/deny decision is automated, ask how they prevent disparate impact, what their adverse-action process is, and whether they'll indemnify you. Watch how fast the room gets quiet. The safe architecture is simpler: screening data assembled by software, decision made by a human — you.

7. "When your AI moves my money, can I cancel it? For how long?" Instant, irreversible money movement by software is a dispute generator. There should be a window — ours is 15 minutes, with a notification the moment anything is queued — between "the AI decided" and "the money moved."

8. "Can I set autonomy per action type — automate this, draft-for-review that, turn off the other?" Trust is earned incrementally. A serious product lets you start conservative and expand; an all-or-nothing switch tells you the product was designed for the demo, not the relationship.

9. "When your AI acts on my behalf, what record exists — and can anyone, including you, edit it afterward?" Landlord disputes are decided on records: when the notice was sent, when the leak was reported, what the deduction was for. A normal database entry is editable and therefore contestable. Ask whether the audit log is tamper-evident — ours hash-chains every entry so any alteration is visible — and whether you can export it. If the vendor's log can be quietly rewritten, so can your evidence.

10. "How does the system handle my state's specific rules — notice periods, deposit deadlines, late-fee caps, required disclosures?" "We have templates" is not an answer. Ask which statute governs security deposit returns in your state and what deadline the system enforces. If they can't answer for your state on the call, the compliance layer is marketing.

11. "When my state's law changes, how fast does the product change, and how will I know?" Landlord-tenant law moves constantly — deposit caps, notice periods, and fee rules changed in multiple states in the last two years alone. Ask for the process: who tracks changes, what the update SLA is, whether updates are logged.

12. "Walk me through your worst recent failure: what the AI got wrong, how it was caught, and what changed." Every AI product has failures. Vendors who claim otherwise are lying or not looking. What you're buying is the failure-handling system: detection, reversal, record, fix. A vendor who answers this question specifically and unprompted is showing you their engineering culture. A vendor who deflects is showing you theirs.

Twelve questions, what a pass looks like, and the red flag to walk away from.

This checklist is free to use — copy it into your own vendor evaluation.

These questions won't tell you whether the product's AI is good — only whether it's safe to trust while you find out. Run the free trial with autonomy set conservative, watch the log for a month, and expand from there. That's how we'd want you to adopt ours.

Key takeaways

Won't every vendor just claim to pass these?

Claims are why every question specifies in writing and in the live product. The written forbidden-actions list in particular is a commitment vendors can't casually make — it constrains their roadmap.

Aren't these questions designed so only Aptoria passes?

They're designed so only products built like Aptoria pass — hard floor, undo window, tamper-evident records. We published our architecture precisely so it can be copied; if these questions become the industry's standard exam, landlords win either way.

What if I'm happy with my current software and just want the AI features?

Then you want a copilot, and that's fine — as long as you're pricing it as one. This checklist is for delegating execution, which is a different decision.

Aptoria