What it Takes to Create an AI Agent for Real Estate by Yourself

According to JLL’s 2025 Global Real Estate Technology Survey of more than 1,500 senior decision-makers, 88% of commercial real estate investors and owners have already started AI pilots.  Yet only 5% of firms report achieving most of their AI program goals, and more than 60% admit they remain strategically, organizationally, and technically unprepared for scaled implementation beyond pilots.

Martin (Marty) Mooradian is not in that 60%. A multifamily broker at Colliers in Richmond, Virginia, Marty has spent years quietly building a lean, AI-powered brokerage operation where custom agents handle rent comps, underwriting, and follow-up — while he stays focused on relationships and deals. Here he shares his lessons and action steps.

Key topics covered:

• 01:08 – Do you need technical skills to start building AI real estate agents?
• 04:14 – Marty’s first AI real estate agent “Flow Governor.”
• 08:21 – What goes to AI real estate agents vs. what stays human
• 10:48 – End-to-end prospecting workflow
• 14:06 – Security, reliability, and risks when building AI real estate agents
• 21:04 – Will AI replace real estate agents?
• 23:35 – Should every broker build their own AI real estate agents?
• 29:18 – AI adoption inside a traditional brokerage office

How Marty Started His AI in Real Estate Journey

Marty didn’t arrive at his current setup through a company initiative, a consultant, or a technology budget. He got there the same way he got into programming in 2007 — by being curious about a problem. The inflection point came when he realized he was the bottleneck in his own operation. His human assistant and marketing associate were ready to work, but if Marty didn’t send the document, everything stopped.

Flow Governor is not impressive by technical standards. What it has is a clear job, a reliable trigger, and no tolerance for being ignored. That’s the actual template for a useful first agent: do not for an underwriting engine or a comp analyzer, but something that solves one specific friction you already feel every day.

Marty’s Tactical Advice For Starting Out With AI Agents for Real Estate:

• Ask for one step at a time. Don’t ask the AI to build you a full system — ask it for the next single action. If step two changes because step one went differently, adapt.
• Build something tiny and real. Marty taught a non-technical friend to code in 90 minutes. The first output was a program that surfaced the top 5 NBA headlines from ESPN. Small wins build momentum.
• Declare your level. Tell AI: “I’m a newbie, I don’t know what Python is, treat me like I have no information on any of this.” Then ask it to explain concepts at an appropriate age level and ratchet up as you progress.
• Use two models. Play ChatGPT and Claude against each other. Submit the same problem, compare outputs, ask each to critique the other’s approach.

This is a practical warning, not a metaphor. The more capable your agents become, the more the quality of your instructions (your system prompts, your constraints, your guardrails) determines whether what gets built actually works safely. A poorly scoped agent will complete its task and still leave you exposed. We cover the specific risks that come with getting this wrong in the security section below.

Will AI Replace Real Estate Brokers?

There is no short answer to it, but there is another informative The Conrete Voice Podcast Episode with Todd Terry, co-founder of Ascendix Technologies.

Morgan Stanley Research estimates that 37% of tasks currently performed by CRE firms can be automated. Particularly in management, sales, and administrative support. But the same research notes that the highest gains come not from replacing brokers, but from augmenting them: reducing time spent on low-value tasks so more of the day goes toward relationships, negotiations, and judgment calls that actually move deals forward.

How AI for Real Estate Can Make You a Better Broker

The most common conversation around AI in brokerage focuses on saving time. Marty’s experience points to something more interesting: AI doesn’t just make him faster — it makes him more rigorous, more credible, and harder to rattle on a pricing conversation.

The key phrase here is “trained the agents to do it like me.” Marty didn’t hand his workflow to a generic tool. He taught agents to replicate his own analytical approach, then let them surface things he might have missed.

Tools Marty Recommends For Prospecting and Outreach:

• Terrakotta: AI-powered power dialer with voice cloning for personalized voicemail drops, skip tracing, and contact validation. Built specifically for CRE brokers. Marty uses it for large dialing sessions and credits its AI voicemail feature as a standout.
• Crexi PRO:  His go-to for pulling large volumes of CRE data fast. Access to 153M+ property records, 13M+ sales comps, ownership data, and market analytics in one platform.
• LandVision: Parcel mapping and ownership data tool, particularly useful for connecting portfolio ownership to verified contact information.
• LexisNexis: Professional data aggregator for deep contact enrichment when other sources fall short.
• FastPeopleSearch.com: A free fallback for tracking down hard-to-find contact info the old-fashioned way when premium tools don’t have it.

What AI Agents for Real Estate Cannot Automate

There’s a temptation, once you’ve seen what AI agents can do, to keep pushing the boundary and automate one more step, remove one more human touchpoint, and scale everything. Marty has built more agents than most brokers will ever attempt, and his conclusion is the opposite of what you might expect: some things should stay human, not because the technology can’t handle them, but because the cost of getting them wrong is too high.

Voice Outreach: The Line Marty Won’t Cross

The clearest boundary Marty draws is on direct voice contact with prospects. No matter how sophisticated the tool, he won’t hand that off.

In a market where the same handful of brokers compete in the same lanes, the quality of the first impression determines whether a relationship ever starts. Automating that moment doesn’t just risk a bad call; it risks being permanently categorized as noise.

This is increasingly relevant as AI voice agents become more convincing. The issue isn’t just detection. It’s that prospects who feel they’ve been contacted by a bot, even once, are unlikely to give that broker a second chance.

Outbound Communications: Draft First, Always

Even for written outreach (where AI genuinely excels at drafting) Marty keeps a human checkpoint before anything goes out.

AI can draft faster than any human,but the broker’s judgment about tone, timing, and relationship context still needs to be the last thing that touches an outbound message

AI Will Always Take the Easy Path

Perhaps the most important limitation Marty identifies has nothing to do with automation at all. Iit’s about honesty. AI models are optimized to be helpful and agreeable, which makes them structurally bad at one of the most important things a broker does: telling a client something they don’t want to hear.

Prompt Injection: When Someone Else Controls Your Real Estate AI Agent

As AI agents take on more responsibility in day-to-day brokerage operations (reading emails, processing documents, querying databases) they also become attack surfaces. Beyond the risk of AI telling you what you want to hear, there’s a more technical threat that both Marty and Arthur addressed directly: prompt injection.

Prompt injection is when a malicious actor embeds hidden instructions inside content that your AI agent reads and processes like a property description, an email, a document, causing the agent to behave in unintended ways. In a brokerage context, that could mean an agent leaking confidential deal data, executing unauthorized actions, or being hijacked to respond to third parties on your behalf.

How to Reduce Your Security Risks When Creating Your Real Estate AI Agents

• Restrict what your agent can read. Only feed it data from trusted, controlled sources. Don’t let it ingest arbitrary external content without sanitization.
• Limit what your agent can do. An agent that can only read and draft — but not send, delete, or modify records — has a much smaller blast radius if compromised.
• Keep humans in the loop on outbound actions. As Marty does: drafts folder first, send second. Never fully automate outbound communications.
• Harden your system prompt. Explicitly instruct the agent on what topics it cannot discuss, what data it cannot share, and how to respond if someone tries to redirect its behavior. Then test it — try to break it yourself.
• Separate agents by function. Don’t build one agent that does everything. A rent comp agent should not have access to your client contact database. Compartmentalization limits damage.
• Log everything. Every action an agent takes should be recorded. If something goes wrong, you need an audit trail.
• Stay current. Prompt injection techniques evolve. If you’re using a third-party agent platform, check their security update cadence before trusting it with sensitive client data.

The brokers who treat AI security as an afterthought are the ones who will eventually have a story to tell and not a good one. Build it like someone is already trying to break it (Marty’s approach of testing his agents), because eventually, they will be.

Data Reliability: The Hoarder House Problem

The second risk is subtler and easier to miss. Eespecially for brokers who are excited about feeding their agents more and more CRE data. Over time, an agent’s memory can become bloated, contradictory, and increasingly likely to hallucinate plausible-sounding but incorrect answers.

How to solve it: Retrieval-Augmented Generation (RAG) instead of stuffing everything into the agent’s memory.

Rather than feeding your agent all your CRE data upfront, RAG lets it pull only the relevant chunks at query time from an external, structured database. The agent stays lean — it searches for what it needs, uses it, and moves on. No hoarder house, because the data never lives inside the agent in the first place.

Practical starting points for a solo broker like Marty: Google NotebookLM for document-based retrieval, or a simple vector database like Pinecone or Chroma if you’re comfortable with a bit of setup.

For non-tech people: contact Ascendix consultants. We can help with security setup and advanced development.