What this session covers
Think building AI agents for your properties is complex and time-consuming? ProptechOS already has 42 agents ready to deploy, covering 23% of the 289 processes typical in real estate and facility management — from leakage monitoring to indoor climate review. In this session, we show you how to get them up and running in just 24 hours, with no IT risk and no change management.
This is Session 1 of 6 in the Agentic Proptech Webinar Series — a practical, hands-on series designed to help real estate and FM professionals understand, build, and scale AI agents across their portfolios.
In this session, we walk through:
- What agentic AI is and how it differs from generative AI tools like ChatGPT or Copilot
- The four types of AI agents used in real estate
- A live build of a stuck damper detection agent using real building data
- How to turn a one-time analysis into an automated agent that runs every day — without human input
Key takeaways
- 37% of real estate tasks can be automated today — not in 5 years. This is Morgan Stanley’s estimate, on top of the 50% efficiency gains already captured by digital tools.
- Agentic AI acts; generative AI answers. Tools like ChatGPT reduce work from hours to minutes. Agents eliminate that work entirely — they run overnight, on weekends, without prompting.
- You don’t need IT to deploy an agent. ProptechOS agents are defined in plain language. You describe what the agent should do, and it reasons from there.
- Start with monitoring, not control. The safest first agents only observe data and report findings. Control agents (e.g., adjusting EV charger load) come once you’ve validated agent behavior.
- A stuck damper agent found real faults in minutes. The live demo scanned 24 air handling units, flagged dampers that moved less than 2 percentage points over 7 days, and produced a reasoned fault report — in one session.
- All ProptechOS agent prompts are open-source on GitHub. You can inspect, adapt, or contribute to any of the 42 defined agents.
- Agents speak your language. Agent prompts work in Danish, Swedish, German, Ukrainian, and more — and they understand context, not just keywords.
The four types of AI Agents in Real Estate
Understanding which agent type fits which task is the foundation of building a scalable agent team for your portfolio.
1. Oracle Agents – The generalist
An oracle has access to all your building data and can answer broad questions across your portfolio — which tenants might churn, where the biggest energy waste is, what happened last Tuesday in Building 4. This is the most familiar type, closest to ChatGPT or Gemini, but connected to your operational data.
2. Expert Agents – The specialist
An expert agent is built for one specific task or process. It doesn’t need to know everything about your portfolio — it needs to know exactly how to gather, analyze, and act on data for one job. An example: an agent that only looks at district heating return temperatures. It won’t reason about leases or occupancy. It will, however, flag every anomaly in return temperature with precision and consistency, every time it runs. Expert agents are the most common worker in a real agent team.
3. Embodied Agents
AI with a body. An embodied agent represents a physical thing — a building, a system, a floor. It is connected to that asset and acts as its voice. A typical first deployment at ProptechOS is one or two building agents that run a health check every morning: are there data anomalies? Has there been an abnormal volume of alerts? Unlike a robot (also an embodied agent), these are stationary — but they know exactly what their building is doing.
4. Task Runners
The fast lane. As your agent team scales, efficiency becomes critical. Task runners are lightweight, cheap, and single-purpose. They can’t reason deeply, but they run constantly — checking conditions, routing exceptions, and handing off to expert agents when something needs escalation. Think of them as the first filter in your automation pipeline.
Practical takeaway: For most FM and real estate teams, the Expert and Task Runner combination covers the vast majority of automated processes. Start there.
What we built live in this session
Stuck Damper Detection Agent
Stuck dampers are a common, expensive, and often invisible problem in HVAC systems. A damper stuck at a fixed position wastes energy continuously and is easy to miss in standard monitoring.
In this session, Per built a stuck damper detection agent live from scratch using real ProptechOS building data:
- Data scanned: 24 variable air volume (VAV) air handling units
- Detection logic: Dampers that vary by less than ±2 percentage points over a 7-day window are flagged as likely stuck
- Output: A reasoned fault report identifying flagged units, with supporting data and severity context
- Build time: Under 15 minutes from blank prompt to running agent
The agent was then saved as a repeatable, scheduled agent — running daily with no human interaction, generating alerts only when a fault condition is detected.
Why not just set a BMS alarm? BMS alarms can flag a condition — but they can’t reason about it. An agent can cross-reference historical data, consider building occupancy patterns, compare linked sensors, and escalate with context. It can also act outside the BMS: generate a work order, send a notification, or update a dashboard automatically.
The ProptechOS Agent ecosystem
ProptechOS sits between your agents and your building systems. It connects to:
- BMS (Building Management Systems)
- Facility management platforms
- ERP systems
- Tenant and lease management tools
- IoT sensor networks
All data is onboarded into the Real Estate Core ontology — an open-source global standard that gives every sensor, room, and system a consistent name across your entire portfolio. This is what makes agents possible at scale: an agent looking for “presence detector in Room 201” finds the right data whether the building is in Copenhagen, Kyiv, or Frankfurt.
The 55+ third-party apps in the ProptechOS ecosystem — covering elevator monitoring, energy optimization, and more — connect into the same layer. Agents can work across apps and systems without custom integration.
All 42 ProptechOS agent prompts are publicly available on GitHub. You can view exactly how each agent is structured — role and identity, context and objectives, constraints and safety rules, and output format. The same library also includes an agent prompt guideline for building your own.
Frequently Asked Questions
Do I need an IT team to deploy these agents?
No. ProptechOS agents are defined in plain language — you describe what the agent should do, it reasons from there. You don’t need to write code or manage infrastructure. Most first agents are running within 24 hours.
How consistent are agent results? Will they give different answers each time?
Agents may express findings differently across runs, but the conclusions they reach are consistent. The agent prompt guidelines include specific instructions for how agents should reason, classify information, and format output — this is what keeps results reliable.
Can agents control building systems, or just monitor them?
Both — but you choose. Monitoring agents observe and report. Control agents (for example, reducing EV charger load when energy prices spike at night) can act on systems. We recommend starting with monitoring agents while you build familiarity and confidence, then expanding to control.
What happens with GDPR and data privacy?
ProptechOS is built with GDPR compliance as a foundation. Agents operate on building operational data (sensors, systems, HVAC) rather than personal data in most use cases. Governance, permissions, and privacy policies are covered in detail in Session 6 of this series.
Can I share an agent I’ve built with colleagues at other organizations?
Yes. Because agent prompts are skill-based rather than data-specific, an agent you build for district heating monitoring in your portfolio will work out of the box for a peer in another city or company. The data it needs to find and how it reasons about it is in the prompt — not hardcoded to your building.
Are these agents only for large portfolios?
No. The agents are designed to work at any scale. A single-building operator can deploy a building health agent. A large municipality can orchestrate dozens of expert agents across hundreds of buildings.
Where can I see all available agent prompts?
All 42 ProptechOS agent prompts, plus the agent prompt guideline and expert specification templates, are publicly available on GitHub. A direct link is available via the QR code shown in this session.
See AI Agents running in your buildings
Book a demo and we can see how you can benefit from AI Agents in your portfolio.
Don’t miss our upcoming webinars in the series, sign up here.
