The future of property and development from local AI to residential ecosystems

At Padel Living Residences in Jeddah, the specification already includes a dedicated app for booking the planned rooftop courts. The project is off-plan, so this is a commitment about how residents will use the building. It also means the booking software will need an operator, support and a budget long after the apartments have been sold.

At a different scale, Wadi Zaha in Sultan Haitham City offers homes facing the planned city’s Central Park. Looking across projects like these in our portfolio, I think the industry needs to give much more attention to the systems and people that will run them after handover. A buyer is making a long-term commitment to a place whose management arrangements will affect everyday life.

My view is that developers will bring more of this work onto a single resident platform, from rent and service-charge administration to maintenance, access and bookings. For those that retain management or service contracts, that creates a continuing commercial relationship with the household. Each additional service then has an existing route to its customer.

Emaar gives an indication of the scale this can reach. In its 2025 annual report, Emaar Development reported more than 98,000 unique active users of Emaar One and over AED 7.2 billion in payments processed through the app. It also described the app as its main channel for community service requests.

In Britain, Ballymore’s b.life combines maintenance tickets and payments with fitness classes, event bookings and contact with the estate team. A resident has reasons to use it throughout the week, beyond paying a bill or reporting a fault.

Live Aldar’s Marketplace takes that relationship into home services, offering furniture packages, maintenance plans and smart home security, among other services. Provision can involve Aldar, linked partners or named third parties, with availability depending on eligibility. A developer can assemble an ecosystem around its residents without having to deliver every service itself.

The limitation is that the resident still has to organise much of the work. They find the right service, fill in the request and explain the history, sometimes to several different people. I expect AI to take on more of that coordination, using records the operator already holds. And I think a growing share of that processing will happen inside the development itself.

A local large language model, or LLM, runs on hardware at the property. That could mean a compact computer in a villa or a shared server serving an apartment building. Connected to the management software, it could handle requests using the building’s own records. Each household would have separate access, and the operator would decide which actions the system could carry out and which needed approval.

Recent model releases make this worth considering in a development brief. Google released Gemma 4 12B in June 2026 with native image and audio inputs alongside language processing. Google says it can run locally with 16GB of video memory or unified memory. Its Apache 2.0 licence also allows businesses to build on the model. Multimodal AI is moving within reach of hardware that could sensibly sit inside a property.

Microsoft’s Phi Silica illustrates another part of the shift. This small language model handles tasks such as summarising and rewriting text on the device. Microsoft states that prompts and model outputs are not transmitted to Microsoft or third parties during inference. Local processing offers a concrete privacy benefit for services that may need to work with information about someone’s home.

For a development, the useful capability would come from connecting a model to accurate records and authorised actions. Consider a resident reporting that the air conditioning has failed again. The system could retrieve the equipment details and repair history, attach a photograph supplied by the resident and prepare a work order for the correct contractor. A facilities manager would receive a case with context, rather than spend time reconstructing it across separate systems.

The model would need to retrieve the relevant information when asked, rather than be trained afresh on every property. That distinction matters commercially: equipment, contractors and residents change. The building needs a reliable, current source of information that can survive a change of model or management company.

The connection to home systems is already being explored. Home Assistant’s Ollama integration allows a locally hosted model to interact with selected devices, although its documentation describes control as experimental. I see this as evidence of the direction of travel. A supported installation across hundreds of homes would still require considerable integration and testing, with conventional building controls retaining responsibility for essential functions.

For international owners, I would expect the service to become especially useful between visits. One request could establish what work has been completed, identify outstanding approvals and arrange cleaning before arrival. Within the permissions given, the same system could help residents book facilities or organise home services. The value would lie in completing those tasks with less coordination from the owner.

I favour a hybrid approach. Routine requests and sensitive household information could be processed locally, with more demanding work passed to cloud services when authorised. Models could be updated as the wider AI sector advances. Keeping inference local reduces the need to transmit information, but the application still needs encryption, separate access rights for each household and clear records of who authorised an action. An operator should have no automatic right to see a resident’s private conversations simply because it manages the building.

For developers, I see an opportunity to extend customer lifetime value through services people continue to choose after handover. Maintenance subscriptions, home preparation and concierge services could create recurring revenues where the developer or its operating partners have the right to provide them. A resident platform supplies distribution; local AI could make those services easier to discover, arrange and use.

The economics could improve when demand is concentrated in one development. A contractor attending several homes on the same estate may spend less time travelling between jobs. Better equipment records could reduce repeat visits or help teams arrive with the right parts. Those are operating benefits worth testing. They also give a developer a reason to invest in the records and integrations behind the resident app.

The same thinking applies to lifestyle. With the resident’s permission, a system could find a free court, arrange a regular booking or alert them to an activity they have asked about. The operator could see which facilities are used and where demand exceeds capacity. Over time, I would expect that information to influence staffing, opening hours and the amenities specified in the next development.

The commercial benefit will differ by tenure. A build-to-rent operator can assess whether better service helps retain tenants and reduces the cost of managing each occupied home. A developer selling homes may see the benefit through its service business, referrals and repeat purchases. None of those outcomes follows automatically from installing AI. They need to be measured against service uptake, repeat complaints, resolution times and the full cost of operating the system.

I would therefore put equipment records, maintenance histories and connections between systems into the development brief before handover. The operating budget needs to cover hardware replacement, software support and security updates. A locally hosted model still has a cost, even where there is no cloud fee for each request. Developers will need to compare the total cost of ownership with the work actually saved.

At Omnia, I believe the quality of this operating model should become part of how we assess a residential development. Buyers are committing to years of ownership. For someone purchasing abroad, confidence in how the home will be maintained and how easily they can use it is a substantial part of the proposition. A persuasive demonstration should therefore be followed by an explanation of the charges, support arrangements and what happens if the provider changes.

I expect the resident app to remain the point of contact while local models take on more of the work behind it. A development could gain new capabilities as those models improve, provided its records and systems can support them. That is the transition I would plan for now. The hardware can be replaced. Reconstructing years of missing maintenance records and fragmented service arrangements will be much harder.