Stop Treating AI Like Software: How to Train AI for Affordable Housing

You probably do not need another AI tool.

You may need a completely different way of thinking about AI.

That is the argument Kent Fai He makes in this solo episode of the Affordable Housing & Real Estate Investing Podcast. Instead of asking which AI model is better or buying another piece of software, Kent asks affordable housing professionals to think about something they already know how to do:

How would you train a new employee?

If you have spent years developing affordable housing, managing properties, raising money, working with lenders, running a nonprofit, or managing a team, you already know how to explain a job, delegate work, review it, and correct mistakes.

Kent believes that same management experience can become the foundation for training AI.

For affordable housing professionals, this matters because our industry is full of repetitive knowledge work. We answer emails. We track grant requirements. We review invoices. We prepare reports. We follow checklists. We move information between systems.

If AI can take even part of that administrative load off someone's plate, Kent's goal is simple: give affordable housing professionals back five or ten hours a week that can be spent building housing and helping people.

How Should Affordable Housing Professionals Think About AI?

Kent starts with a message for people who feel like the AI world is moving faster than they can keep up:

“You're not behind, you're just busy.”

Most experienced professionals have spent the last five, ten, or fifteen years doing their actual jobs. They have been serving clients, closing deals, managing projects, raising money, solving problems, and keeping organizations running.

Meanwhile, much of the AI conversation has centered on which model is better.

Kent thinks that misses the bigger opportunity.

His core idea is:

Stop thinking about AI as software you need to learn. Start thinking about AI as someone you can delegate work to.

That changes the question.

Instead of asking, “Which AI tool should I buy?” ask:

What work would I hand to a very fast new employee who has almost no context about my organization?

A new employee may be intelligent and capable, but you would not expect that person to understand your organization's history, standards, customers, financing structures, communication style, or unwritten rules on day two.

AI has the same problem.

It needs context.

It needs examples.

It needs instructions.

It needs feedback.

And it needs to know when not to make a decision.

What Is Kent Fai He's Define, Delegate, Decide AI Framework?

Kent introduces what he calls the 3Ds: Define, Delegate, Decide.

It is a simple framework for thinking about AI implementation without getting lost in technical language.

1. Define

First, define what good looks like.

What is the goal?

What information does the AI need?

What rules should it follow?

What does an acceptable result look like?

What should it never do?

This is essentially the same thing a good manager does before assigning work.

2. Delegate

Next, hand off the repeatable execution.

Kent connects this to his 10/80/10 framework from the previous episode.

The first 10% is the human defining the goal, standards, and context.

The middle 80% is where AI can potentially perform the repeatable work.

The final 10% returns to the human.

3. Decide

You review what AI produced.

Is it correct?

Does it sound like you?

Did it follow the rules?

Should it actually be used?

The human still decides what goes out the door.

That is an important distinction. Kent is not describing a system where AI independently runs an affordable housing organization. He is describing delegation with human oversight.

How Can AI Help With Affordable Housing Workflows?

Kent uses email as a simple example because almost every organization has a “front door.”

For a developer, it might be a new business inquiry.

For a housing organization, someone may ask whether units are available or how to apply.

For a foundation, it might be a donor asking how to support the mission or a grantee asking about the next funding cycle.

The industries are different, but Kent argues that many incoming questions repeat.

The first exercise does not even require opening an AI tool.

Go through the last 30 to 60 days of your inbox.

Do not document every email. Instead, identify the types of emails you repeatedly receive.

Then divide them into three buckets.

Bucket 1: Routine

These are questions where the answer is largely the same each time.

A property manager might receive:

“Do you have units available?”

“What is the application fee?”

“How do I apply?”

A foundation may repeatedly answer when its next grant cycle opens.

These are strong candidates for AI assisted drafting because the rules and responses can be clearly documented.

Bucket 2: Draft, Then Review

These questions have a real answer, but circumstances matter.

AI can use past examples and your organization's playbook to prepare a draft, but somebody should review it before anything happens.

Bucket 3: Escalate to a Human

This is where Kent draws a hard line.

Legal threats, eviction related communications, sensitive resident circumstances, allegations of unfair treatment, or other high risk situations should not simply flow through an automated response system.

Sometimes the correct AI action is:

Stop and get a human.

That is part of good automation too.

How Can AI Help Affordable Housing Asset Management and Compliance?

Kent gives an affordable housing specific example involving invoices and grants.

Imagine a multifamily project with numerous funding sources.

One grant might reimburse only qualifying labor. Another funding source could have specific requirements around eligible materials. Staff then need to review invoices, separate costs, document compliance, and track which expenses were paid from which source.

People perform this type of work every day.

AI could potentially assist with pieces of that process when the rules are clearly documented.

The key word is clearly.

You cannot simply tell an AI system to “review the invoice.”

You need to explain what it is looking for, which funding rules apply, what documentation is required, what should be flagged, and when a person needs to review the result.

That level of detail is what turns generic AI into something customized to an affordable housing organization's actual workflow.

It also explains one of Kent's central concepts:

“Whenever you have a problem, it's really not an AI problem, it's an instructions problem.”

What Are AI Agents and AI Skills?

The terminology around AI can make relatively simple ideas sound intimidating.

Kent strips it down.

An agent has a goal or role.

A skill gives that agent instructions for completing a particular task.

Kent compares a skill to a document sitting inside a labeled filing cabinet.

Imagine an AI agent responsible for marketing. It might need one skill explaining how to research topics and another explaining how your organization writes LinkedIn posts.

The agent does not need every instruction for every task loaded all the time. It needs to know which instructions apply to the job in front of it.

For affordable housing, the same idea could apply to:

• Preparing recurring funder reports

• Reviewing project information

• Drafting routine stakeholder communications

• Following an internal underwriting checklist

• Processing recurring property management inquiries

• Organizing grant compliance information

The real asset is not the fancy terminology.

It is the documented know how behind the task.

Your checklists, templates, examples, corrections, rules, and institutional experience become instructions an AI system can reference.

How Do You Train AI Without Losing Human Judgment?

Kent's approach requires guardrails.

One of the most important instructions you can give an AI system is:

If you are unsure, do not guess.

Kent points out that a confident wrong answer can be much worse than no answer at all.

That is why AI should be trained like a new employee.

If someone starts working for you on Monday, would you expect that person to understand everything about your organization by Tuesday?

Probably not.

You give them feedback.

You correct mistakes.

You explain why two situations that look similar actually require different responses.

Kent recommends doing the same thing with AI.

If the system drafts an email, preserve the draft. Then compare it with the email you actually send.

What did you change?

Did you change the tone?

Did it misunderstand the type of inquiry?

Did it include something you would never say?

That correction becomes another piece of training.

This leads to one of the most powerful concepts in the episode.

When an experienced employee leaves, years of knowledge can walk out the door with them.

When an organization documents its standards, examples, rules, and corrections for AI, that knowledge can become part of the organization's intellectual property.

You are not simply training AI.

You are finally documenting how your organization actually works.

Key Insights and Frameworks

• Define, Delegate, Decide: Define success and context, delegate appropriate execution to AI, then decide whether the output meets your standards.

• 10/80/10 Framework: Humans handle the first 10% by setting the standard, AI can assist with the middle 80% of execution, and humans handle the final 10% through review and judgment.

• Three Bucket Framework: Separate incoming work into routine tasks, tasks that AI can draft but humans should review, and sensitive tasks that should immediately escalate to a person.

• Instructions are intellectual property: Your SOPs, templates, examples, rules, and corrections capture years of organizational experience.

• Treat AI like a new hire: Do not expect perfection immediately. Train it, review its work, correct mistakes, and document what it learns.

Best Quotes From Kent Fai He

“You're not behind, you're just busy.”

“Stop thinking about AI as software.”

“Whenever you have a problem, it's really not an AI problem, it's an instructions problem.”

“A confident wrong answer, it's just so much worse than no answer at all.”

“Treat it like human, not software.”

Why Could AI Matter So Much for Affordable Housing?

Affordable housing professionals already deal with complexity.

Projects can involve multiple funding sources, compliance requirements, reporting obligations, lenders, government programs, residents, consultants, and countless documents.

AI does not eliminate that complexity.

But it may reduce the amount of human time spent repeatedly moving information through it.

Kent's vision is not to remove people from affordable housing.

It is almost the opposite.

Use AI for repeatable administrative work so people have more time for the parts of housing that require human judgment, relationships, creativity, empathy, negotiation, and leadership.

As Kent explains near the end of the episode, if AI can give someone back five or ten hours every week, that means more time and energy can go toward housing and helping others.

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.

Common Questions This Episode Answers

How do I train AI for my business?

Start by defining the tasks you repeatedly perform, then document what good work looks like. Give AI examples, rules, instructions, and clear escalation criteria, then review and correct its work just as you would with a new employee.

What is the Define, Delegate, Decide framework?

Define means establishing the goal, context, and standard. Delegate means letting AI perform appropriate repeatable work, while Decide means bringing human judgment back in to review and approve the result.

What is Kent Fai He's 10/80/10 AI framework?

The first 10% is human preparation and definition. AI handles much of the middle 80% of execution, then a human handles the final 10% through review, correction, and decision making.

What affordable housing tasks could AI help automate?

Kent discusses areas such as routine emails, property management inquiries, grant and invoice tracking, compliance workflows, and other tasks with clear rules. Sensitive legal, resident, collection, or eviction related matters require greater caution and human involvement.

What is an AI skill?

Kent describes an AI skill as essentially a set of instructions used when an agent needs to perform a particular task. The important part is documenting the SOP, examples, rules, and expected output clearly enough that AI knows what to do.

Stop Asking Whether AI Will Work for You

The biggest lesson from this episode is not about a particular AI model.

It is about management.

If you know how to train somebody, explain your standards, delegate work, review results, and give feedback, you already understand much of the thinking required to train AI.

The harder work is getting the knowledge out of your head.

Write down the rules.

Document the exceptions.

Save examples.

Explain what you never do.

Define when a human needs to step in.

Then improve those instructions every time the system gets something wrong.

The next episode will go deeper into the actual software Kent uses and how these ideas can be implemented for specific businesses.

For affordable housing professionals, the goal remains bigger than AI itself.

Get administrative time back so we can spend more of it creating housing and helping people.

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.