Haven Insiders · {{current_date_full}}
What we shipped, what we're thinking about, and what's moving in property management.
FROM MADELINE
What does your AI actually know?
Spotify offered to make me a podcast this week.
Not a playlist. A podcast, generated on demand, to prep me for an upcoming meeting. Their new Studio feature (early exploration, and credit to them for experimenting) takes a prompt like “help me prepare for a meeting” and produces a chatty little episode about it. Lifelong Spotify user here, and I love some of their AI features. So I tried it on my weekly sync with my boss. What came back was polished, confident, and useless, because Spotify does not know my product, my roadmap, or anything about the person across the table. It knows my running playlists.
I posted about it on LinkedIn this morning, and the point is worth repeating here: to get good outputs, you need context. An agent overlaid on a music platform is not equipped to create meeting prep podcasts. Whatever a meeting prep podcast is. (Regular readers will recognize last issue’s theme too: restraint is a feature, and knowing what not to build is part of it.)
Now bring that home to our industry. Two property management firms buy the same AI tool. Same model, same features, same price. Six months later one of them swears by it and the other canceled in week three.
The difference is almost never the AI. It is what the AI knew.
I grew up in my family’s property management business, and I keep thinking about what a first week looked like for a new coordinator. They were smart. They were motivated. They were also useless, through no fault of their own. They did not know that the owner on Elm Street approves anything under $500 without a call. They did not know which plumber actually shows up on Saturdays and which one just says he will. They did not know that the “leak” in unit 12 is condensation off an old line, every single summer.
Nobody expects a new hire to perform without that. We hand them SOPs, we let them shadow, we answer four hundred questions in their first month. Then we hand an AI a login and expect it to be brilliant on day one with none of the above.
The models are not the bottleneck anymore. Every vendor in this industry has access to roughly the same intelligence, and it is getting cheaper by the quarter. What separates a useful agent from a generic one is context: what it knows about your properties, your owners, your vendors, your rules. An agent that knows your emergency criteria triages a 10pm call differently than one guessing at industry averages. An agent that knows an owner’s approval threshold closes the loop instead of adding a step to it.
Here is the part that surprised me. More context is not the goal either. I spent part of this month auditing where our agents get their knowledge, and then doing the same audit on my own AI setup. On the personal side I had connected everything to everything, hundreds of tools and data sources, and my assistant got dumber the more I added. When I cut it back to the handful that mattered, it got sharp again. Noise is not context. A pile of stale documents is not context. Context is the smallest set of true, current facts about your operation, in front of the right agent at the right moment.
That is a data problem, and data problems are unglamorous. Cleaning up vendor lists. Writing down the rules that live in one coordinator’s head. Deciding which system is telling the truth when two of them disagree. Nobody got into property management because they love this work. But it is quietly becoming the line between firms where AI compounds and firms where it stalls.
We think about this constantly at Haven. A surprising share of our roadmap is not “smarter model.” It is making sure your agents know what your best coordinator knows. The intelligence is table stakes. The knowledge is yours, and it is worth more than you think.
If you do one thing this week: write down three rules about your operation that exist only in someone’s head. That is context. Whatever AI you point at your business next will be better for it.
This week’s essay, in one image (seen on ProgrammerHumor.io):
PRODUCT
Product Updates & Releases
New
The Work Orders page, rebuilt. The Maintenance Conversations tab got the full do-over treatment. It is now a proper Work Orders page: search, filters, and a clean detail view for every job, organized around the work itself instead of the text thread.
Everything you used the old tab for is still there. It is just faster to find and easier to act on. Firms on our newer maintenance experience have it now; everyone else gets it as that experience rolls out.
Talk to your agents in notes. This one is a big deal. Leave a plain-English note on a work order, “do not respond,” “already handled,” and Haven reads it and acts on it: it holds off, wraps up, or stands down. No settings screen, no toggle, no ticket to us. You write to the agent the way you would write to a coordinator, and it listens. As of this week that works for every firm, automatically. (Yes, this was last issue’s pro tip. It earned a promotion.)
Smarter call handoffs. When a Receptionist call needs a person, the handoff now runs through a new, more reliable transfer flow, and whoever picks up hears who is calling before they connect. Fewer fumbled transfers at the exact moment a caller has the least patience.
Improved
Photo submissions from residents are recognized more reliably during maintenance requests.
Haven is smarter about old work orders: an edit to an already-resolved job no longer triggers unnecessary check-in texts, and residents will not get double-messaged about an issue that already has an open work order.
Bugs, squashed
A bug where residents who texted STOP could still get follow-up messages. Gone, fully.
A bug that sent vendors an incorrect “reassigned” message when a job was accidentally double-booked.
A bug where status-update and vendor-assignment texts only went out for some firms. They now go out for every firm, every time.
Two call-transfer bugs: one that could loop a call instead of connecting it, and one that could drop a caller instead of holding them during a handoff.
In flight
This is the part where I get excited about things you cannot click yet.
A unified Settings Hub. Haven settings are currently a bit of a scavenger hunt, and we’ve been working hard to make thing easier to navigate! The new Settings hub brings permissions, users, calendars, firm details, and notifications into one place, and yes, it will finally let you invite your own team yourself. Truly groundbreaking innovation, we know. It is a big build and it is moving fast.
Vendor coordination. Haven working directly with your vendors on a work order: the outreach, the scheduling back-and-forth, the follow-ups your coordinators handle by phone today. Not just relaying through your team. This is the one I am most excited about.
A rebuilt leasing inbox. Filters instead of tabs, bulk actions, and one fast place to run every prospect conversation.
Resident responsibility settings. Tell Haven which issues are on the resident and which are on the firm, and let it pull the terms straight from the lease. Fewer “actually, the garbage disposal is on you” conversations for your team to have.
More on the way, stay tuned.
EARLY ACCESS
Want Haven to schedule the visit?
Scheduling is the newest agent in Haven Maintenance, and we are opening the next round of pilot spots. Here is what it does: the moment you assign a vendor in your PMS, Haven opens a private text thread between the vendor and the resident, listens for a time that works for both, and writes the confirmed appointment back to the work order. No app, no login, no one on your team playing message relay. If the vendor backs out, Haven catches it and flags you. If the resident texts that the issue fixed itself, Haven confirms and closes it out. It runs on Buildium, Rentvine, and AppFolio.
One note on fit. This round is built for firms whose maintenance mostly goes to third-party vendors. A few in-house techs are fine; there is a setting for that. If you run primarily on in-house technicians, scheduling for your team deserves its own build, not a retrofit, and it is on our list. You will read about it here first.
If that sounds like your operation, put your name in. It takes about thirty seconds, and we will follow up with onboarding.
ON OUR RADAR
Things I've Bookmarked Recently
Effective Context Engineering for AI Agents (Anthropic): the team behind Claude on why stuffing more into an AI’s memory backfires (they call it “context rot”), and why the real skill is choosing the smallest set of high-signal facts. The technical companion to this week’s essay.
Your Data Agents Need Context (a16z): why an agent that has your raw data but not your definitions will be confidently wrong. The plainest business-side explanation of garbage in, garbage out I have found.
Why 95% of AI Pilots Fail (Forbes, on MIT’s research): the models are not the problem. Most pilots never get connected to real workflows and real data. Worth reading before your next tool evaluation.
PRO TIP
Tips & Tricks
Your onboarding answers are your agents’ brain. The property notes, vendor preferences, and emergency criteria you shared when you set up Haven are exactly what your agents reason with on every conversation. If your vendor bench or after-hours rules have changed since onboarding, tell us. Five minutes of updates makes every future conversation smarter.
SHARE
Know someone who'd like this?
If a fellow property management pro would get something out of Haven Insiders, send them your personal link below. Thanks for spreading the word.
Go give your agents something true to work with.
Here's to the Builders,
Madeline
How was this issue?