What the AI does
PMFriend uses AI for language tasks — things that are hard to do with rules but easy for a person, where the only reason a human is typing is because nobody's automated it yet.
The eight places AI helps
| Surface | What the AI does | What a human still does |
|---|---|---|
| Maintenance triage | Reads the tenant's description, suggests category + urgency + likely causes | PM accepts, edits, or overrides the suggestion |
| Duplicate detection | Spots when the same issue's been reported twice | PM confirms before merging or dismissing |
| Work-order scope | Drafts a scope brief from the issue + AU rental context | PM reviews, sets cost ceiling, picks the contractor |
| Contractor matching | Ranks contractors by trade, recency, insurance, rating | PM picks anyone they want from the full list |
| Owner digest | Drafts a fortnightly update from actual property activity | PM reads, edits, sends |
| Inspection report | Writes 1-2 page prose from the room-by-room checklist | PM reviews, edits, sends |
| Legal notices | Drafts arrears + entry notices with state-specific form references | PM verifies, edits, sends |
| Tribunal case packs | Assembles chronology + Statement of Facts + relief paragraph | PM verifies, signs off, takes to tribunal |
What the AI never does
The AI never:
- Sends anything to anyone autonomously. Nothing reaches an owner, tenant, or contractor without a property manager clicking a button.
- Makes legally significant decisions. It doesn't decide whether a notice is valid, whether a tenancy is breached, or whether evidence is strong enough for tribunal. That's all human judgment.
- Has access to your trust ledger. Trust accounting lives in your PMS. PMFriend doesn't read or write to it.
- Trains on your data. We use AI through commercial APIs that contractually don't train on customer data. Your inspection reports aren't going into anyone else's training set.
What "AI" means in this context
When we say AI, we mean large language models — the kind that read text and write text. PMFriend uses commercial AI providers (with Australian data-residency where available) to handle these tasks.
We don't use AI for:
- Numerical calculations (those are deterministic code)
- Scheduling logic (rules-based)
- Permission checks (rules-based)
- The data layer (a regular Postgres database in Sydney)
Think of the AI as a very fast typist who's read every PM textbook in print, can write in AU English, and never gets tired. It's not a decision-maker. It's a typing-and-summarising assistant.