Agentic AI in Real Estate: 4 Steps to Train It Like a New Hire

Most real estate agents have crossed the first bridge with AI. They use it to write listing descriptions, clean up an email, or answer a question they used to google.

The next move is bigger. Agentic AI connects to the software you already use and performs the work, while chat only hands you words to copy and paste.

REDX Academy lead trainer Tyler Fenn and REDX president Curtis Fenn spent a full episode on this, and their advice runs opposite to instinct. The work starts at a whiteboard. Write down what you actually do every day, then decide what you would hand to an assistant. Here are the four steps they lay out, plus the mistake that eats entire afternoons.

What Is Agentic AI in Real Estate?

Agentic AI is a model with permission to act inside your tools. You can give it access to your inbox, browser, design software, or CRM, and it carries out multi-step work on your behalf.

The current crop of these tools includes ChatGPT agent mode, Claude Desktop with Cowork, Grok Bot, Manus, Codex, and Perplexity Comet. Most start around $20 a month.

The connection layer has a few names that all mean roughly the same thing: connectors, integrations, MCPs. “If you don’t know what that is, it doesn’t matter,” Curtis says. “Just ask the AI what you need to do to get it connected.”

A handful of MLSs have already published these connections, along with the mainstream tools most agents already live in.

Why Do So Many Agents Stay Busy With AI and Get Nothing Out of It?

Because activity inside a tool feels like progress. It is the biggest trap waiting for agents who are new to AI. Curtis wrote about this pattern in Inman recently, and his comparison is to the early days of social media.

“If people aren’t focused in the right areas, it gives the illusion that they’re being productive without actually doing anything. You can learn to do a lot and never accomplish anything if you’re not careful.”

Tyler’s example is a friend who announced he had finally jumped in on AI and spent six hours building automations. Six hours of building, zero conversations with sellers. Despite working really hard for 6 hours, nothing came of it.

So to fix this first issue, you need to set a guardrail before you open the tool. Decide what you will finish, decide when you are shutting it down, and go back to the activity that produces appointments.

What Should You Do Before You Open an AI Tool?

Write down what you already do. Most agents cannot delegate their business because it has never left their head.

Step 1: Document one ordinary day

At the end of a normal workday, list everything you did. Then look for what repeats every day and every week: checking email a dozen times, pulling the MLS on Monday to see what came on and off the market over the weekend, preparing the same snapshot for the same kind of seller.

Tyler tests this constantly in training. “Do you have a documented series of steps that you follow when you get a referral? Nine out of 10 agents are like, no, I just know what I do.”

Step 2: Run the 10-assistant test

Now ignore AI completely and ask a staffing question: if you had 10 assistants tomorrow, what would you hand them? Keep the list to work you are doing today.

The trap here is inventing new projects. “If you’re not doing it now, then don’t distract yourself with trying to do it yet,” Curtis says. The goal of this round is space on your calendar, and you get that by removing existing work.

David Van Noy, a team leader and former Mike Ferry coach, hit this same fork while he was training his own assistants. “I realized I had an opportunity at that point to either train more people or train the machine,” Van Noy said on REDX Monday Motivation LIVE.

Van Noy sorts the list with one rule before any task moves to an assistant or to an AI. He splits the work in his business into two categories and gives each category an owner.

“One is personality, the other one is perfection. … Anything that requires perfection should be done by the AI. Anything that requires personality should contain you.”

Our guide on what real estate virtual assistants can take off your plate covers the same delegation question from the staffing side.

Step 3: Write the training steps

Pick one item off that list and write out how you would train a new virtual assistant to do it: every click, every rule, every exception. That document is what you hand the AI, and it is why people who start here get results in an afternoon while others spend a week prompting.

Van Noy made the same case for writing the process down. “I bet I’ve trained 40 assistants in the last 10 years, and what I would try to get them to do is press the same button in the exact same way,” he said. A written SOP is how that same button gets pressed the same way every time.

How Do You Train Agentic AI Like an Employee?

Treat it as staff, the same as anyone else on your team. Tyler says that single reframe changes how agents use these tools and what they get back from them.

A new hire arrives with context: a job description, interviews, maybe prior experience doing the work. Your AI arrives with the average knowledge of the internet and nothing about your business, so you onboard it in the same order you would onboard a person:

1. Give it context. Here is my business, my market, my role, your job, and what good output looks like.
2. Give it tool access. Connect the inbox, the browser, the design tool, the CRM, whatever the task requires.
3. Set permission levels. Decide what it may do alone and what needs your approval.
4. Feed it the SOP you wrote in step 3 and tell it to take over that task.
5. Run a dry run. Have it walk the whole process, produce the output, and tell you where the output will land.
6. Close the feedback loop after every real attempt.

Permissions are where agents get burned

Be explicit about the line between drafting and sending. The permission Tyler sets first:

Never send mail from my inbox. Draft it and wait for my approval.

You would set that same boundary with a brand new assistant on day one.

The most important feedback loop most people skip

Here is the habit that separates a tool you tolerate from an assistant that improves. Most agents take an AI draft, decide it is a decent first pass, paste it somewhere else, edit it, and publish. That edit is the training data, and it gets thrown away.

Bring the finished version back instead. “This is the final product that I used. Compare the two and ask me questions about why I did it the way I did it.”

A new employee who did a task wrong would get that same correction on the spot, before being sent to do it again.

How Do You Move From Single Tasks to Real Automation?

Hand it outcomes and let it work out the steps. Once the AI knows several of your processes, Curtis says the level up is managing it like a smart employee: tell it what you want to be true, and let it figure out the mechanics.

His first one was email. The outcome he wanted was simple: stop being interrupted all day, and tell me what matters at 10:00 and again at 2:00. He never specified how to sort, filter, or summarize anything.

The prompt he still uses to pull the details out of his own head:

“Ask me between five and 10 questions to clarify.”

The tool then asks how you want the report, how you want to be notified, what counts as urgent, and what to ignore. Answer its questions, then drop back into the details only when it gets stuck. 

Our breakdown of AI versus automation for real estate agents covers where the line sits between a prompt and a real workflow.

Van Noy has already carried this past a single task. By his count, his transaction process once had 146 steps handled by a person.

“I found in my system there was 146 items that were essentially handled by a human inside of transactions. As of today, there’s four of them that are handled by me and 46 that are handled by one assistant, which means we displaced over 100 activities with AI, and my objective is to get that all the way down to maybe 10 or 12.”

Van Noy hands every agent the same filter for their own list. “Every task that you have currently … it does one of two things. It requires either judgment, which the AI cannot replace, or repetition in anything that you do the same way twice.”

Which Agentic AI Tool Should a Real Estate Agent Start With?

Pick one for about $20 a month and start talking to it. Curtis suggests Claude Desktop first, with a steeper learning curve, and Manus as the follow-up because it is the most user friendly on-ramp from chat to action. ChatGPT agent mode, Codex, Perplexity, and Gemini all have versions of this.

Setup solves itself in a way earlier software never managed. Ask the tool to walk you through connecting it, and it will tell you it needs access to your email to do what you asked, then that it needs your browser to log into the MLS for you, and it will say plainly what it cannot do.

“This is the first time in the history of our species where we have a tool available to us and the tool itself can tell us how to use it,” Tyler says. “We don’t have to go somewhere else to learn.”

Carry one caution: this category changes monthly, so treat any tool recommendation, including this one, as a snapshot. Ask your AI what is current before you commit a workflow to it. Our guide to automating your first recurring task gives you a narrower starting point if prospecting is where you want help first.

Your First Assignment This Week

Set aside 90 minutes, and spend the first 30 of them away from a keyboard.

1. List your repeating work. Everything you did today, then circle what happens daily or weekly.
2. Choose the most boring item on the list. High frequency, low judgment, no client relationship attached.
3. Write the training steps as if a new assistant starts Monday and has never seen your business.
4. Subscribe to one agentic tool and paste that document in with a plain question: can you do this for me, and what access do you need?
5. Set your permissions out loud. Name what it may never do without your approval, starting with anything that reaches a client.
6. Dry run it, then correct it once. Give it your edited version and have it ask you why you changed what you changed.
7. Bank the time you got back into listing appointments, then repeat with the next item.

Keep the last step honest. This business still runs on trust between two people in a kitchen, and Curtis’s framing at the end of the conversation is the right test for every automation you build: use AI to take over the work you are not required for, and spend the hours you recover face to face with the people who will hire you.

Key Takeaways

• Agentic AI acts inside your tools. Chat produces text you paste somewhere; an agent with access to your inbox, browser, and CRM completes the task.
• Start on paper. Document one ordinary day, run the 10-assistant test, and write the training steps for one repeating task before you open any tool.
• Onboard it like a hire. Context, tool access, permission levels, your SOP, a dry run, then a feedback loop where you show it your edited final version.
• Assign outcomes once it knows your processes. Ask it to clarify with five to 10 questions, and give the recovered hours back to seller conversations.