This is not the future. This is running in production now.
37% of all tasks in the real estate industry are automatable with today’s technology. Not in three years. Not after your data is clean. Now.
This opening session of the Agentic Proptech Webinar Series shows what agentic building operations actually look like in practice — with live demos, two customer case studies, and a working team of AI agents running across a real portfolio in real time.
Rasmus and Per walk through how generative AI and agentic AI differ, what a team of AI agents looks like inside ProptechOS, and how to create and deploy your first autonomous agent — live, on screen, in under five minutes. Howard Hughes Corporation shares how they moved from pilot to production across lease administration, ERP abstraction, and operational workflows. Locum shares how a task that previously took two days of external consultant work now takes four hours.
The question is no longer whether this will matter. The question is how far behind you want to be.
What this session covers
- Why 37% is probably an underestimate — the first 50% of efficiency gains took 40 years across the digital era; the next wave will happen in three to five years, and the pace is already accelerating beyond initial projections
- Generative AI vs. agentic AI — the difference between an assistant that answers questions and a digital colleague that acts, reasons, and works autonomously every hour of every day
- How AI agent teams are structured — embodied building agents, expert agents with specific mandates, and task runners that do simple work continuously and escalate when needed
- Live demo: creating an autonomous agent — a nighttime ventilation saver agent deployed across a portfolio, running every night at 2am, finding 290 MWh per year in energy waste on its first run — saving €60,000 annually with no ongoing human effort
- Live demo: multi-agent collaboration — a peak shaving workflow where a task runner detects rising demand, an energy expert proposes a layered load-shedding plan, the building agent evaluates it, escalates to a human for approval, and a supervisor agent executes — all within minutes
- Howard Hughes Corporation — how one of the US’s largest mixed-use real estate companies moved past pilots into production, cutting lease drafting from weeks to minutes and abstracting executed leases into their ERP automatically
- Locum — how the real estate company managing Stockholm’s hospitals and care facilities reduced a two-day external consultant task to four hours using agentic workflows
- 292 processes mapped — how ProptechOS identified every process in a real estate organisation, found that only six are purely manual, and built 42 expert agents covering 23% of them today
What the live demo shows
Creating an autonomous agent from scratch
A nighttime ventilation saver agent is built live during the webinar. The system prompt defines the agent’s objective: detect ventilation energy waste during unoccupied hours, correlate airflow and presence data, comply with ISO and ASHRAE standards, and generate a prioritised waste report every night at 2am. Permission policy is assigned — portfolio-wide read access. A schedule trigger is set. The agent is live.
On its first real-world run for a ProptechOS customer, this agent found 290 MWh per year in wasted energy — €60,000 annually. It also detected a faulty presence sensor mid-analysis and, without any instruction to do so, located a working CO2 sensor nearby and used it to infer occupancy instead. No code. No rule. The agent reasoned its way to a solution.
Multi-agent peak shaving
A task runner monitors power consumption continuously. When load approaches a critical threshold, it escalates to the peak shaving expert agent. The expert analyses the situation, throttles EV chargers outside the building to shed 17 kW, and monitors whether that is sufficient. It is not. The agent escalates to the building agent — the AI representing the physical building — which evaluates the situation, proposes lowering corridor heating to 17°C for an additional 20 kW reduction, and flags the decision for human approval. A human approves. A supervisor agent executes the plan. Three agents. One human touchpoint. Zero dashboards to monitor.
Complaint triage and enrichment
A task runner monitors incoming FM tickets. When a temperature complaint arrives — “the office landscape alpha is too cold; people are working in their winter coats” — it escalates to the service object triage expert. The expert pulls historical data, analyses adjacent rooms, correlates with BMS data, and prepares a full enriched brief before any human needs to look at it. What used to require manual investigation across multiple systems is ready before the facility manager opens their inbox.
What customers say
Marcus Baini, Senior Vice President, Howard Hughes Corporation
Howard Hughes manages large-scale mixed-use communities across the United States — retail, office, hospitality, multifamily, and master-planned communities. 80% of their innovation team’s work is now focused on AI.
Lease drafting: a first draft lease, customised to the tenant’s region and local law, now takes minutes. It previously took days or weeks of external legal work even when working from a template. Once a lease is executed, it is automatically abstracted — common area maintenance, recharges, and all relevant detail pushed into their ERP system with a human in a review position. Weeks of elapsed time reduced to minutes.
“It isn’t a question anymore of if this will matter. It’s how far ahead do you want to be. Data is messy. Just get started. Agentic workflows can now handle imperfect data and improve as they go. Waiting for the perfect platform means you are always waiting.”
Thomas Ahlberg, Locum (Stockholm Region Real Estate)
Locum manages the hospitals, care facilities, and acute care buildings of the Stockholm region — among the most complex and operationally critical buildings in the world. A task that previously required two days of external consultant work now takes four hours. Thomas described it as the biggest and best change in his working life in the sector — and as genuinely fun to work with.
How ProptechOS AI agents are structured
Every agent in the ProptechOS agentic suite has three components:
System prompt
The agent’s job description: what it monitors, what it should prioritise, what standards it must comply with, what it should do when it finds something, and when to escalate to another agent or a human. Agents can be instructed in any language — English, Swedish, Norwegian, Danish, or a mix. No programming knowledge required.
Permission policy
What the agent is allowed to do is separate from what it is instructed to do. Permissions are watertight. An agent cannot access data or tools outside its assigned scope. A monitoring agent can read sensor data across a portfolio but cannot change a set point. An executive agent can implement approved plans but only within the buildings and systems it has been explicitly granted access to. Agents do not decide their own permissions — those are set by the operator.
Triggers
What causes the agent to act. This can be a timer — every night at 2am — an event in the building data, a threshold being crossed, or a message from another agent. Task runners monitor continuously and call in expert agents when needed. Expert agents escalate to building agents or humans when a decision exceeds their authority.
The 42 agents available today
ProptechOS has mapped 292 unique processes across a real estate organisation. Only six are purely manual. 54% are either largely or fully automatable today. The current agent suite — 42 expert agents — covers 23% of all identified processes and grows continuously.
Agents are grouped into three types:
Embodied building agents — AI representing a specific physical building, with access to that building’s data only, responsible for its operational health and acting as a coordinator for other agents working within it.
Expert agents — specialised AI with deep knowledge of a specific domain: energy management, ventilation optimisation, lease administration, complaint triage, alarm consolidation, predictive maintenance, data quality monitoring, and more.
Task runners — lightweight models that run continuously, perform simple monitoring and analysis, and escalate to expert agents when conditions are met. Low cost, always on, never fatigued.
No-brainer starting points for any organisation: ticket triage, alarm consolidation across systems, and context enrichment for incoming work orders. These require no changes to existing systems, carry no data regulation risk, and eliminate work that currently falls through the gaps every day.
Frequently asked questions
Q: Do I need clean data before I can start?
No. Howard Hughes — one of the largest and most sophisticated real estate companies in the US — started with imperfect data. Agentic AI has evolved to handle messy, inconsistent data and improve as it goes. Waiting for perfect data means you are always waiting. The recommendation from every customer in this webinar is the same: get started.
Q: Do I need to know how to code or write in English?
No. Agents are configured through system prompts written in plain language. Any language works — Swedish, Norwegian, Danish, German, or a mix. The AI understands the instructions regardless of which language they are written in.
Q: What is the difference between generative AI and agentic AI in ProptechOS?
Generative AI in ProptechOS is the built-in assistant — ask a question, get an answer, generate a report, analyse a dataset. You are always in the loop and always initiating. Agentic AI is a separate layer: autonomous agents with their own identity, their own permissions, and their own schedules. They work without being asked. They escalate when they need a decision. They act within their permissions without human initiation.
Q: How are agents kept safe — what stops them doing something they should not?
Permissions and instructions are strictly separated. An agent can be instructed to optimise energy across a portfolio, but its permission policy defines exactly which buildings and systems it can touch and what actions it can take. Instructions tell the agent what to aim for. Permissions define the hard boundaries of what it can actually do. These cannot be overridden by the agent itself.
Q: If I am already a ProptechOS customer, can I access the agents now?
Yes. If ProptechOS is already running in your buildings, the agent suite is available to you now. The agents use the same data and tools as the built-in AI assistant — just with their own identity, their own permissions, and autonomous operation.
Q: What if I am not a customer yet?
ProptechOS can help you identify the right use cases and user stories to start with based on your portfolio, your systems, and your operational priorities — before any technical commitment is required.
See AI agents working in your buildings
Book a demo and we can walk through which agents make sense for your portfolio and what you could have running within days.
