Seven Domains of Multifamily Software and the Two Kinds of AI
Updated: Oct 30
AI that works for the application makes that vendor stickier. AI that works for the user, reaching in through another vendor's API and MCP server, makes software headless. Neither has redrawn the seven domains yet. The vendor that holds the core object still sets the terms.
On October 7, 2026, RealPage paused new activations of EliseAI, a leasing AI company. The pause is an early public sign of who controls which AI agents get into a property management system. As of October 27, 2026, I found no announcement ending it.
This is the multifamily example in a four-part series on agentic impact on core systems. Application Equilibrium set out the theory: every domain is built around one core object. A domain holds its shape until a core forms around an object that sat inside an existing core.
A core object is the record or definition around which a product organizes its primary process and data model, such as the lease and the ledger or the key performance indicator. A core is the product built around that object. Its vendor sets the terms on which anyone else can reach it.
Here I apply the theory to multifamily, where I have identified seven domains. The property management system (PMS) holds one of them.
I drew this map, and the call on each domain is mine. Will AI Agents Replace Systems of Record? then asks what changes when agents become the users.
Key question. Will AI create a new domain in multifamily software, or only strengthen the seven it has?
Key finding. Multifamily software has seven domains, each built around its own core object. AI strengthens the vendor that holds each one, and nothing so far shows it creating a new core object. Outside agents, which make the software headless, reach that object on the holder's terms.
Agents Are Becoming Users of Multifamily Software, and Two Kinds of AI Act on It
Agents are becoming users of multifamily software. A headless system keeps its data and its rules, and the screen becomes optional. Agents read and write it through an API, the interface programmers use, and a Model Context Protocol (MCP) server, a standard interface built for agents. I call the pair the door.
Salesforce pitched exactly that in April 2026.¹ Ed Sim of boldstart ventures describes the incumbent stance as "have at it, agents, have at my platform. You're just going to have to pay per access."² RealPage CEO Dirk Wakeham puts it this way: "Users might not even log into systems anymore."³
Two kinds of AI act on this software, and they do different things.
Internal AI works for the application. It stays inside one vendor's boundary and improves that vendor's domain on that vendor's data: a leasing assistant answering from the customer record, a pricing model built on the KPIs, a maintenance tool reading the condition history. The bet: it makes the vendor stickier.
If AI improves the vendor's domain, reliance on the vendor grows, and every improvement raises the cost of leaving.
External AI works for the user. It is an agent bound to a person's role, not to any one product, and it is external to the system it enters.
It reaches into each vendor's system through the door as a source and a tool: it reads a rent roll from the PMS or, where the vendor permits writes, posts a charge to the ledger. It is what makes software headless.
The kind depends on the system the agent enters. It does not depend on who built it. An agent that reaches into another vendor's system through the door is external to that system, whoever built it. So the vendors that hold core objects will build both.
EliseAI's Apollo, the agent RealPage says was never submitted for review, is the case on this map. It is EliseAI's own agent, external to the RealPage system it reaches into.
I make three points:
Multifamily software has seven domains, each built around its own core object.
Entrants and vendors extend those domains or grow into a neighbor, and none has created a new core object.
AI strengthens whoever holds the object, while outside agents meet the core's terms.
Multifamily Software Has Seven Domains, Each Built Around Its Own Core Object
Multifamily software sorts into seven domains. Each is organized around a single core object, a record or definition that no other domain is built around. Many vendors keep records about the same object. Many compete inside each domain. What counts is which product is built around the object, because the rest of the domain depends on that product.
Here is the map. Each row names the domain, the object it is organized around, and a representative player.
Domain | Core object | Representative player(s) |
Shopping and Selecting | The listing | Zillow, Apartments.com |
Customers and Communications | The customer across a portfolio, as prospect, applicant, resident and former resident | Funnel, EliseAI |
Leases and Ledgers | The lease and the ledger | Yardi, MRI Software, RealPage, Entrata, AppFolio, ResMan |
Repairs and Routines | The building and the unit, with their condition history | HappyCo |
Data and Decisions | The key performance indicator (KPI) | REBA |
People and Performance | The worker | Horizontal HR platforms such as Workday |
Deals and Documents | The deal | AcquiOS, the closest attempt; no clear holder |
The objects are not all the same kind of thing. Some are records of entities. Some are compound records. One is a shared definition. The test is the same across them: which record or definition organizes the product's primary process and data model? A product can consume or reproduce that object without becoming the core built around it.
Shopping and Selecting sits upstream of everything else, before a prospect is even a lead. Customers and Communications, Leases and Ledgers, Repairs and Routines and Data and Decisions are the four every operator deals with daily.
People and Performance runs horizontal to the rest. Deals and Documents belongs to a different buyer entirely, the asset manager's stack rather than the property's.
Three of the other six came out of the PMS. Customers and Communications, Repairs and Routines and Data and Decisions each started from an element of the PMS's data: the prospect, the work order and the reporting. Each became a different object: the customer, the building and unit, and the KPI. Leases and Ledgers is the domain they broke from.
The other three did not start as PMS data. Shopping and Selecting grew out of advertising, where software automated the link between a listing and the channel that carries it. The worker and the deal began outside the PMS too.
Shopping and Selecting: the object is the listing
This domain is where a prospect compares options before they're a lead in anyone's customer relationship management (CRM) system. Listing price, availability, photos, syndicated across the portals a search turns up.
Apartments.com and Zillow are the entrants that matter here. In February 2025 Zillow and Redfin entered an agreement over rental advertising. The Federal Trade Commission (FTC) sued over it in September 2025. In August 2026 it announced a proposed order resolving the suit: syndication continues without the restraints.⁴
Neither company holds the lease, the ledger or the customer record across a portfolio. They hold the listing, and increasingly, who gets to see it at all.
Zillow is already using AI to defend its hold on the listing rather than dilute it. Its AI Mode, which debuted in beta in March 2026, answers a renter's questions and lets them schedule a tour inside Zillow's own experience.⁵
Its Gemini integration, announced in July 2026, goes further. A renter can search inside Google's own assistant, but the tour is still Zillow's. It sits on Zillow's listing and shows in the renter's Zillow itinerary, whether Gemini submits it, as Google's help page describes, or the renter clicks through, as Inman reports.⁶
That is external AI meeting a core object's terms. Google's assistant gets the search. Zillow keeps the booking record.
Customers and Communications: the object is the customer across a portfolio
This domain is the CRM and leasing front end, and the portals on both sides of a signed lease. Prospect inquiry, tour scheduling, application on the applicant side. Rent payment, renewal, work order submission, package tracking, resident communication on the other. The core object is the customer, in four states: prospect, applicant, resident and former resident.
This is most of what a resident actually experiences: find, tour, sign, move in, live there. The lease itself is the one moment that hands off to Leases and Ledgers. It is the most crowded domain in this list. Conversational AI is only one part of it.
EliseAI says one in six apartments in the United States now runs on its products.⁷ EliseAI's pages describe a single customer record per person across a portfolio,⁷ and Funnel, a multifamily CRM company, describes the same on its page.⁸
Leases and Ledgers: the object is the lease and the ledger
This domain is the property management system. Accounting, the lease, the general ledger, trust accounting, and the resident's own ledger, charges, payments, balance owed, tied to the lease but its own record. It carries the compliance weight: a CFO's signature depends on what this system says is true.
The major PMS platforms built their businesses on this domain, and their AI here is internal. It is straightforward automation of what the ledger already does, such as reading invoices, auditing leases for missed charges and helping with the close.⁹
It's exactly what you'd expect an established core to do to improve itself. It's also the domain with almost no outside entrant at any tier so far. That fits everything else here.
Repairs and Routines: the object is the building and the unit
This domain covers maintenance and operations. Work orders, unit turns, inspections, vendor sourcing. The object is the building and the unit as a record of condition, every inspection, repair and turn written against them. A work order is an event on that record, not the record itself.
HappyCo says it serves more than 6 million units, and its press release counted more than a billion maintenance interactions, integrated with the major PMS platforms.¹⁰ The history compounds with every work order. Anything built on it draws on whatever its holder has recorded longest.
Data and Decisions: the object is the KPI
This domain covers business intelligence, pricing and forecasting. Its object is the KPI: the agreed definition of a measure such as occupancy, effective rent or net operating income, defined once across every core a portfolio runs. IBM calls the place those definitions live a semantic layer and says it provides a single source of truth.¹¹
The domain consumes what the cores produce. The definitions that make them comparable are what it holds. A pricing recommendation is an output built on the KPIs, not the object itself.
REBA unifies operational, financial and leasing data into a single analytics platform. It connects to the data sources a client already runs. Its named AI layer, rebaAI, runs on top of that same data. It analyzes supply, demand and performance trends to generate pricing recommendations. It is built on data REBA already holds, not on a new claim of its own.
Its site shows Brookfield Properties, Mill Creek Residential and Pretium among its clients,¹² which suggests it has cleared the referenceability bar, the proof of named clients a buyer can call.
That bar stops most challengers before a first sale. Most customers do not have time to co-develop software, so references are the proof points, hopefully unbiased, that a solution works.
People and Performance: the object is the worker
This domain covers the staff running all of this. Scheduling, task assignment, training, the human side of operations rather than the asset side. This one runs horizontal to the other four the way payroll and HR run horizontal across industries, not native to multifamily specifically.
Horizontal HR platforms such as Workday hold the worker object, and few in proptech care about the gap.
It is also the one domain whose object is a person. The other six are things an operator manages, and the worker is the one doing the managing.
Deals and Documents: the object is the deal
This domain covers acquisition and disposition. Underwriting, diligence, closing. This is the asset manager's domain, not the property's. Its object is the deal itself, a transaction that has to happen before any of the other five have anything to manage: no lease, no resident, no work order until someone has already bought the building.
The object is the deal record: the screening history, underwriting, diligence file and closing documents for one transaction, kept together. Modeling tools and document management serve that record without being it.
AcquiOS goes further, saying it maintains a searchable deal registry of every property a team has screened,¹³ the domain's closest attempt at holding the deal as its own object. A firm's own screening history is a dataset no other domain produces. That is what makes a registry of it worth holding.
Entrants and Vendors Extend a Domain or Grow Into a Neighbor, and None Has Created a New Core Object
None of the three ways in, described below, works today without the door. An entrant that can't read from and write back to the PMS can't extend a domain deeper. It can only sit next to the system of record and hope someone copies data over by hand.
The door, the API and MCP server through which an outside agent reaches a system, is the direct way in, and the vendor that holds the core object controls it. RealPage's October pause on EliseAI, described in the next section, shows what that means. The door exists. That vendor sets the terms for getting through it.
To place any vendor, ask three questions in order. Is its object native to one of the seven domains? If not, it is the fourth thing, a horizontal entrant, below.
Is it the PMS, the system built around the lease and the ledger, reaching into another domain? Then the PMS is growing outward.
If it is not the PMS, is it moving toward the lease and the ledger, or adding depth around the object it already holds? The first is moving toward the PMS. The second is extending a domain deeper.
The PMS growing outward. This is how every vendor that holds a core in these seven domains operates. It grows outward, by acquisition, by bundling, by whatever tool is at hand, and AI is just the most recent one.
Major PMS platforms are now shipping AI tools of their own, leasing assistants among them, reaching into Customers and Communications, a domain whose object is the customer, not the lease.⁹
Moving toward the PMS. Entrata did this once, completed years ago: portal and payments, moving inward from Customers and Communications to Leases and Ledgers over the better part of a decade. That was the slow, expensive, referenceability-gated path. It worked because Entrata earned trust at the edge before it ever touched a ledger.
It also had a head start the older incumbents didn't: born cloud-native while they were still built on older infrastructure.
No company on this map has repeated Entrata's route. What companies do now is land in one domain and expand into its neighbors. EliseAI landed in leasing and has moved into CRM, maintenance, delinquency and renewals, and Bilt landed in rent payments and, reportedly, has moved into CRM and leasing workflows.¹⁴
Whether either reaches the ledger is the open question. Entrata's route was a specific historical opening, not a strategy replicable at will. Once the incumbents modernized, that door closed. A report that EliseAI is building its own PMS, covered in the next section, tests whether it is closed.
Extending a domain deeper. This is where most of the field lives. REBA aggregates lease, financial and operational data the PMS already produces into decision support the PMS did not offer.
Much of what Funnel and HappyCo ship belongs here too: tour scheduling, application workflows, work-order routing, vendor sourcing, predictive maintenance, approval routing, payment matching. These tasks run on objects the PMS already tracked in some form: a prospect, a work order, a vendor, an invoice.
The exception worth naming is where a company holds something genuinely new rather than only adding depth. Funnel's customer record is not the default object in the PMS platforms I reviewed: one person tracked across every lease they hold, not a separate row per property. Some PMS platforms can be configured toward something similar.
The distinction is not whether a PMS has customer records. It is whether the product is organized around one customer history that persists across properties, rather than around each property's lease and ledger. That decides whether a renter's history follows them from one property to the next.
A fourth thing, running on a different axis entirely. Payments processing companies like Zego enter multifamily as they enter other verticals. They carry an object that has nothing to do with a lease, a customer, or an asset. A payment doesn't care what it's paying for.
Zego was acquired by Global Payments in a deal announced in 2021. Global Payments framed it as entry into real estate, a $6.5 billion addressable market with a strong payments nexus.¹⁵
A horizontal entrant isn't competing for any of the seven objects. It's routed through all seven at once and holds none. A payment lands in the lease's ledger, so the ledger stays the record and the payments company is a rail into it, not an eighth domain. That is part of the market's structure, not an exception to it.
AI Strengthens Whoever Holds the Object, and Outside Agents Meet the Core's Terms
On October 7, 2026, RealPage paused new EliseAI site activations. Its CEO told Multifamily Dive the reason was an EliseAI agent called Apollo that was never submitted for review. He also cited reports of attempts to work around RealPage's controls and a concern for tenant data protection. He said "We're not trying to box Elise out."
RealPage also cites EliseAI's obligations under the RealPage Exchange agreement. Existing integrations are unaffected. EliseAI had not commented when Multifamily Dive published its story.¹⁶
The pause is a bump in the road to settled terms for agent access, not a closed door.
Internal AI strengthens whoever already holds a domain's core object. When agents are the users, value shifts from owning the screen to controlling who may read and write the object. The vendor that holds the object sets those terms.
Some agents already have access, on terms that vendor sets. Yardi has offered a Claude connector since September 2025. Its June 2026 release says access follows existing user permissions. RealPage opened read-only MCP access to customers' AI tools in August 2026. It says its own AI Leasing Agent is working across more than 175 management companies.¹⁷
RealPage's pause on EliseAI shows the gated end of access: the vendor that holds the core object decides which agents it has reviewed.
One pattern runs through the AI-driven moves on this map. EliseAI's customer record, HappyCo's condition history, the PMS-native assistants, REBA's forecasting: each comes from an incumbent or a company that already holds its object. So far they go deeper into ground they already have, with AI that works on their own data.
EliseAI is the AI-led company on the map, and it holds an existing domain's object, the customer record. It created no new one. That holds even if the report proves true. Thesis Driven reports, from executives who have seen demos, that EliseAI is building an AI-first PMS for an early 2027 launch.¹⁸ EliseAI has not confirmed it. A PMS is built on the lease and the ledger, so EliseAI would hold a second existing object and still create no new one. It would be the first AI-led company on this map to go for the ledger directly.
The external kind, agents working through someone else's door, has produced a negotiation over access, not a new owner. Nothing on this map shows AI creating a new core object. A new core object created on the strength of AI would break it. The AI is new. The advantage it's amplifying isn't.
What an operator should do with this. Before the next renewal, ask each vendor: does an agent I choose get read and write access to my data, what does that access cost, and who decides which agents are approved? My red flag is write access with no record of which agent acted.
What a vendor should do with this. Decide now which agents you will review and on what terms, because RealPage's pause shows that the decision gets made in public.
What an investor should do with this. Watch for an AI-native product whose record is none of the seven objects, and which other vendors then have to build around. That is the one signal that the equilibrium of the seven domains is about to change. If the EliseAI PMS launches, check what it is organized around. My read is that EliseAI keeps the customer record in its CRM and builds the PMS clean around the lease and the ledger. That makes it another instance of an existing domain, with EliseAI holding two domains' objects in two separate products. The check at launch is whether a customer can use one without the other.
A New Core Object That None of the Seven Holds Would Change This Picture
Each row is a judgment about which product the rest of the domain is built around. A different call on one row would change that row, not the test.
Five domains have players built around their object. People and Performance belongs to horizontal HR platforms. Deals and Documents has no clear holder. A newcomer can compete in any of them, so competing is not the test. The test is whether AI produces a core object that none of the seven holds.
A record is none of the seven if it cannot be derived from any of them and other vendors' products have to read or write it through the newcomer.
A second case to watch is an outside agent that coordinates work across several of the cores built around these objects and holds none of the objects. It could become commercially important without becoming a core.
That is the orchestration agent taken up in Will AI Agents Replace Systems of Record?, and it would sharpen the line between an orchestration agent and a core rather than break it.
This map applies the object test from Application Equilibrium to one vertical; Will AI Agents Replace Systems of Record? works it backward from 2032, and What Stays Scarce in a World of Agents follows the person.
There are four seats in the name. The fifth one is yours.
¹ Salesforce, April 21, 2026: 6 New Innovations Redefining Salesforce Development, which says every capability Salesforce has built over 25 years "is now available via API, MCP, or CLI." That post does not use the name Headless 360. Beri, April 20, 2026: Salesforce Headless 360: The UI Becomes Optional, which reports the April 15 launch and the Headless 360 name.
² Ed Sim of boldstart ventures, on CXOTalk, August 9, 2026: AI agents and systems of record.
³ Faraudo, Propmodo, August 26, 2026: AI is quietly reshaping the economics of property management, which attributes the quote to RealPage CEO Dirk Wakeham.
⁴ The FTC's suit over the February 2025 Zillow and Redfin agreement, September 30, 2025: FTC. The August 24, 2026 order: FTC, which calls it a stipulated order that takes legal force when a federal judge approves it. It says Redfin keeps syndicating Zillow's listings without the earlier restraints and must restart its rental listing business within six months of finalization.
⁵ Zillow, March 25, 2026: Zillow debuts AI Mode.
⁶ Inman, July 9, 2026: Zillow rentals in Google Gemini. Google's help page: Gemini and Zillow.
⁷ EliseAI, June 10, 2026: press release. EliseCRM, which describes a "single, living guest profile" for each renter and a sync with the PMS: EliseAI. EliseAI's blog, which argues for a "single guest card per prospect that follows them across every property in the portfolio": EliseAI.
⁹ Yardi, June 15, 2026: Yardi delivers AI agents for every multifamily role and workflow, which lists Chat IQ for lead nurturing, tour scheduling, application support, payment reminders and renewal outreach, Smart AP for invoice data, and agents that audit leases for missed charges and help with month-end close. Yardi Smart AP, October 29, 2025: Yardi, which reads invoices by optical character recognition and sends the data to Breeze Premier for staff to review, approve and post. AppFolio, June 16, 2026: AppFolio, which describes a Leasing Performer that answers inquiries and schedules tours, an Accounting Performer that imports and codes invoices, and a Resident Messenger Performer. RealPage, August 10, 2026: RealPage introduces Lumina, which includes an AI leasing agent. These three are examples. I did not read the other major platforms' releases.
¹⁰ HappyCo: press release and commissioned research, run for HappyCo by Freed Vance Research Group. HappyCo's About page now says more than 6 million units.
¹¹ IBM, What is a semantic layer?: IBM. Metric definitions and a single source of truth are IBM's description of the layer, not a multifamily source.
¹² REBA: getreba.com.
¹⁴ EliseAI's platform page, which lists products for leasing, EliseCRM, maintenance, delinquency and renewals: EliseAI; I read it October 20, 2026. Bilt, January 24, 2025: Resident Loyalty 2.0, which covers leasing incentives and resident engagement and does not mention a CRM. Bilt's move into CRM rests on Hargreaves, Thesis Driven, July 7, 2026: The Great Proptech Convergence.
¹⁵ Global Payments, May 4, 2021: Global Payments to acquire Zego.
¹⁶ RealPage CEO Dirk Wakeham, October 12, 2026: Multifamily Dive. The October 7 pause was first reported by Thesis Driven. The Thesis Driven letter is partly paywalled, and the CEO's statement was read in Multifamily Dive.
¹⁷ Yardi's Virtuoso Connector for Claude, June 16, 2026: Yardi, first announced September 10, 2025: Yardi. RealPage Lumina, August 10, 2026, including the AI leasing agent figure: RealPage. The read-only detail: RealPage Lumina Connect, which says initial Lumina Model Connect access to RealPage data is read-only, user permissioned and authenticated, and works with ChatGPT or Codex, with Claude and Copilot planned.
¹⁸ Thesis Driven, Exclusive: EliseAI Is Building a Property Management System: Thesis Driven. The letter is partly paywalled, and I read its visible text on October 30, 2026. It relies on unnamed executives who have seen demos, and EliseAI has not confirmed the project.
Comments