Who Should Train Your AI? A Real Estate Brokerage Playbook

Most real estate professionals are testing AI the wrong way.

They open ChatGPT, Claude, Gemini, or an AI feature inside their email. They paste in a message and type:

“Write me a reply.”

The AI produces something generic. It sounds robotic. Maybe it gets an important detail wrong.

The agent or broker looks at the result and decides AI does not work.

Kent Fai He argues that the problem may not be the AI at all.

The problem is that nobody trained it.

In this solo episode of the Affordable Housing & Real Estate Investing Podcast, Kent takes the AI frameworks from his previous episodes and applies them to a real business: a real estate brokerage.

Instead of discussing AI in theory, he walks through how a brokerage could train AI to classify incoming emails, draft routine responses, identify risky situations, learn from human corrections, and gradually improve without allowing AI to make decisions it should not make.

The lesson also applies far beyond traditional real estate brokerage. Affordable housing developers, property managers, nonprofits, asset managers, fundraisers, and housing organizations all deal with repetitive communication and institutional knowledge.

The opportunity is not simply to use AI.

It is to build a system where AI gets better because your people get better at training it.

Who Should Train AI Inside a Real Estate Brokerage?

This is one of the most important questions Kent tackles in the episode.

If the owner or broker has to personally train AI on every email, every correction, and every exception, AI could create another job instead of eliminating work.

Kent's solution is to divide the responsibility among three roles.

Role 1: The Broker or Team Leader

The broker defines what good looks like.

They establish the rules, standards, boundaries, and situations that require human judgment.

They also have final authority over whether a proposed new rule should become permanent.

Role 2: The AI

AI does the repetitive work.

It reads an incoming email, determines what type of request it is, and follows the appropriate instructions.

Depending on the situation, it might draft a response, flag something for review, ask a question, or stop completely.

Role 3: The Reviewer

The reviewer is the person closest to the daily workflow.

That might be an inside sales agent, newer agent, transaction coordinator, or another team member who already handles incoming leads.

The reviewer checks the AI's drafts, corrects mistakes, and identifies patterns that may deserve a new permanent rule.

This structure matters because the broker does not need to spend the entire day rewriting AI responses.

Over time, Kent's vision is that the broker could spend a much smaller amount of time reviewing proposed rule changes, while the team handles the daily feedback loop.

That is how AI begins to scale inside a business.

What Is the Define, Delegate, Decide Framework for AI?

Kent brings back his 3Ds framework: Define, Delegate, Decide.

Define

Write down what good looks like.

What are your rules?

How should your company respond?

What language should your team use?

What should AI never say?

What requires a licensed professional?

What situations are too risky to automate?

Delegate

Give AI the work that follows those rules.

For a real estate brokerage, that could mean reading incoming emails, classifying them, and preparing drafts.

Decide

A human checks the work before it goes out.

Kent makes an important distinction here:

You can delegate the work. You cannot delegate the accountability.

For a broker, the brokerage license and professional responsibility do not disappear because AI helped prepare an email.

This is why the framework is more useful than simply asking, “Can AI automate my email?”

The better question is:

Which parts can AI handle, which parts need review, and which parts should remain human?

Which Real Estate Emails Should AI Handle?

Kent recommends dividing incoming communication into three buckets.

This is one of the most practical concepts in the episode.

Bucket 1: Routine

These are repetitive questions that generally receive the same type of answer.

Examples might include:

• Do you cover my area?

• Is this listing still available?

• How can I schedule a showing?

• Where can I find this information?

These are the easiest candidates for AI assisted workflows.

Bucket 2: Draft and Review

These questions have legitimate answers, but context matters.

Someone might ask what their home is worth, about commission, or about another issue where a licensed professional should review what is being communicated.

AI can potentially prepare the first draft.

A human reviews it.

Bucket 3: Human

These are situations where the cost of a mistake is too high.

Kent discusses examples involving wiring instructions, contracts, offers, legal concerns, complaints, fair housing issues, and other sensitive situations.

If AI cannot determine whether something belongs in Bucket 2 or Bucket 3, Kent recommends taking the safer route:

Put it in Bucket 3.

This creates a critical AI principle for housing organizations:

Automation should not mean removing humans from every process. Good automation knows when to bring a human back in.

How Do You Train an AI Agent Without Writing Code?

Kent makes the implementation surprisingly simple.

Before buying more software, go through the last 30 to 60 days of your inbox.

Write down the types of emails you receive.

You do not need to document every message.

Kent estimates a brokerage might discover that eight, ten, or fifteen common types cover a large portion of incoming email.

Then create three simple files.

File 1: The Process

This file explains what AI should do when an email arrives.

For example:

1. Read the email.

2. Determine its type.

3. Assign it to the correct bucket.

4. If it is routine, prepare the approved response.

5. If it needs review, draft a response and explain what the reviewer should check.

6. If it is high risk, do not draft the response. Label it for a human.

7. If the rules conflict or the AI is unsure, stop and ask.

File 2: Email Types and Rules

This file explains the different categories and how to distinguish between similar situations.

For example, someone asking about the brokerage's listing is different from a homeowner asking what their own property is worth.

Those questions may sound similar to an AI system unless the distinction is documented.

File 3: Voice and Guardrails

This contains the words you actually use.

For routine questions, you can provide exact approved responses.

You can also document things the AI should never do.

Kent gives examples such as not providing a home value over email, not presenting commission as a fixed number, being careful about subjective neighborhood descriptions, and treating wiring information with extreme caution.

The point is not that every brokerage should adopt the same rules.

The point is that your rules need to be written down.

Kent emphasizes that these files can be plain text.

No sophisticated programming is required.

“I didn't write a single line of code for my skill files or markdown files. I just wrote down the rules.”

How Do You Stop AI From Giving Confident Wrong Answers?

One of Kent's biggest concerns is not that AI will say “I don't know.”

It is that AI can sound confident when it is wrong.

That means an AI workflow needs explicit instructions for uncertainty.

Teach AI When to Stop and Ask

If two rules appear to conflict, do not guess.

If an email appears to fit multiple categories, do not guess.

If important information is missing, do not guess.

Instead, label the item and ask a human one clear question.

Kent explains the tradeoff well.

An AI system stopping to ask a question may cost somebody 30 seconds.

A confidently wrong email could cost a client.

The same principle is important for affordable housing.

If an AI system is helping review compliance documents, funding requirements, property management communication, or underwriting information, the goal should not be to eliminate every human touch.

The goal should be to eliminate unnecessary human work without eliminating necessary human judgment.

How Can AI Learn From Your Team's Corrections?

This is where Kent's framework becomes more interesting than simple email automation.

Imagine AI writes a draft.

Your reviewer changes three sentences before sending it.

Most businesses stop there.

Kent says the system should learn from the difference between the AI draft and the final human version.

Once a day, the AI could compare:

What did AI originally write?

versus:

What did the human actually send?

Then record what changed.

The reviewer might maintain a simple log containing:

• Date

• Type of email

• What AI wrote

• What the reviewer changed

• Why it changed

• Whether the correction might apply to future emails

Not every correction becomes a permanent rule.

That is where the broker or team leader returns.

Kent suggests that eventually the broker could periodically review only the proposed permanent changes.

Maybe a law changed.

Maybe company policy changed.

Maybe the team discovered that a certain response repeatedly creates confusion.

The broker decides yes or no.

Approved changes go back into the instruction files.

From then on, future drafts follow the updated rule.

Kent calls this a self learning loop.

The system improves because real human feedback becomes reusable organizational knowledge.

The Core Ideas and Concepts From This Episode

1. AI Training Should Be a Team Sport

The owner should not personally correct every AI output forever. Leadership defines standards, AI performs repetitive work, reviewers catch mistakes, and leadership approves lasting rule changes.

2. You Can Delegate Work, Not Accountability

AI may draft or organize work, but the responsible professional still owns the final decision.

3. Teach AI When Not to Act

A mature AI system should know when to draft, when to request review, when to ask a question, and when to stop completely.

4. Every Correction Can Become Institutional Knowledge

Human edits should not disappear after an email is sent. Useful corrections can become documented rules that improve future work.

5. The Goal Is Better Human Work

Kent's vision is not to automate away the most talented people. It is to move them toward negotiation, client relationships, valuation, problem solving, and other work where human experience matters most.

Best Quotes From Kent Fai He

“You can't delegate the accountability.”

“If AI gives you bad work, chances are it was always not an AI problem, it was an instruction problem.”

“I didn't write a single line of code for my skill files or markdown files. I just wrote down the rules.”

“A confident wrong answer can cost you a whole client.”

“Every correction you write, suggest a change or give feedback to an agent, it kind of remembered forever.”

Common Questions This Episode Answers

Who should train AI inside a real estate brokerage?

Kent recommends separating responsibilities. The broker or team leader establishes standards and approves permanent rules, AI handles repeatable execution, and a reviewer checks daily outputs and identifies potential improvements.

How can a real estate brokerage use AI for email?

Start by categorizing the types of emails the brokerage receives. Routine messages may be good candidates for AI assisted drafting, while contextual or higher risk communications should require human review or escalation.

How do you train an AI agent without coding?

Kent recommends documenting the workflow, different request types, approved language, and guardrails in simple files. The focus is on writing down business rules in plain language rather than programming everything from scratch.

How do you prevent AI from making dangerous mistakes?

Give the system explicit stop and ask rules. When the AI encounters conflicting instructions, missing information, sensitive issues, or a situation outside its authority, it should escalate instead of guessing.

Can AI learn from employee corrections?

Kent describes a feedback loop where AI compares its original draft with the final human edited version. Repeated corrections can be proposed as new rules, reviewed by leadership, and added to the organization's permanent instructions.

A 6 Week Roadmap for Training AI in a Real Estate Business

Kent ends the episode with a practical path forward.

Week 1: Review recent emails. Identify the recurring types and divide them into buckets.

Week 2: Write the three files covering your process, email categories, approved responses, voice, and guardrails.

Week 3: Test the system against historical emails. Include examples AI should refuse or escalate.

Weeks 4 through 6: Begin using the workflow on incoming messages while humans continue reviewing drafts and recording corrections.

Over time, those corrections can become better instructions.

The business becomes more efficient, but something else happens too.

Knowledge stops living exclusively inside people's heads.

When an experienced employee leaves a company, years of tribal knowledge can leave with them.

A well designed AI training system forces a company to document what it knows, why it does things a certain way, what mistakes it has encountered, and what standards it expects.

That may eventually become more valuable than the original email automation.

For affordable housing organizations, the same concept could be applied to repeatable administrative processes, asset management workflows, reporting, fundraising, project intake, property management communication, and other areas where clear rules already exist.

The goal is not AI for the sake of AI.

It is exactly what Kent has emphasized throughout this series:

Apply AI to the boring work so talented people can spend more time doing meaningful work.

Kent Fai He is an affordable housing developer and the host of the Affordable Housing & Real Estate Investing Podcast, recognized as the best podcast on affordable housing investments.

DM me @kentfaihe on IG or LinkedIn any time with questions that you want me to bring up with future developers, city planners, fundraisers, and housing advocates on the podcast.

Disclaimer: This content is for informational and entertainment purposes only. It is not legal, financial, investment, insurance, or tax advice. It is not an offer or solicitation for any investments. Always do your own research before making investment decisions.