EP 390 Jev Changes Everything for Real Estate Investors—Except This
APPLE PODCASTS | SPOTIFY
There’s a lot of hype around Jev right now.
And we get it.
The pitch is almost perfectly designed for anyone running a real estate investment company:
Stop asking AI to write paragraphs. Start asking it to make decisions.
If you’re staring at hundreds of offering memorandums, rent rolls, T-12s, broker emails, and deals waiting to be screened, that sounds pretty incredible.
Upload the package.
Ask whether the deal fits.
Get a clean yes or no.
Move on.
Suddenly, the dream of automating acquisitions feels a whole lot closer.
But after building a real estate portfolio approaching $500 million, we’ve learned that finding a deal is rarely the hardest part.
The hard part is maintaining good judgment from the moment a deal enters your inbox to the moment you decide whether to put real capital behind it—and then maintaining that continuity after you own it.
And that’s the part of the Jev conversation we think is getting skipped.
Because ChatGPT can write a beautiful investment memo.
Jev can give you a beautifully structured decision.
Neither one automatically knows whether you asked the right question.
So before we hand acquisitions over to AI, we need to talk about what Jev actually does, what’s hype versus reality, and the missing process that matters regardless of which AI model everyone is talking about next month.
First, What Exactly Is Jev?
Jev isn’t another chatbot.
And that’s important.
Jev is TypeSafe’s first public System One model.
Instead of asking an open-ended question and waiting for paragraphs of text, you define the possible answers ahead of time.
Then Jev evaluates the information and returns a structured decision—with probabilities and confidence attached.
Think:
Yes or no.
Option A, B, or C.
Where does this fall on a scale?
For a multifamily investment company, that could look like:
Does this property meet our minimum unit count?
Which market bucket does it belong in?
Is the deal package complete?
Do the OM and T-12 appear to conflict?
Does this deal deserve a full underwrite today?
Notice how different those questions are from:
“Is this a good deal?”
That’s the point.
Jev is designed to answer narrow, defined questions that software can actually do something with.
And we think that’s genuinely interesting.
Think of Jev Like an Intelligent If-Statement
Here’s probably the simplest way we can explain it.
Imagine an if-statement in software:
IF property has more than 100 units → continue.
IF market is approved → continue.
IF required documents are missing → request them.
Except some investment questions aren’t perfectly deterministic.
You can’t always write:
IF X = 7, THEN Y = TRUE.
Sometimes you’re asking whether two documents appear inconsistent.
Or which risk category best describes a situation.
Or whether something deserves further review.
That’s where a model like Jev becomes interesting.
We think of it as something like an intelligent, probabilistic if-statement.
Chat is optimized to generate language.
Jev is optimized to make constrained judgments that another piece of software can inspect and route.
That’s a real difference.
But—and this is a very important but—
A new kind of AI doesn’t eliminate an old investing problem.
Fast Doesn’t Mean Right
This is where we think the Jev conversation gets especially interesting.
TypeSafe currently lists Jev at 4.2 cents per million input tokens, with no charge for output tokens. Its launch materials also report end-to-end response times around 70 to 500 milliseconds and describe some System One-style queries as dramatically faster than frontier models.
TypeSafe has also published workflow tests showing peaks of 193.6x faster and 444.6x cheaper, while explicitly describing those figures as being at the higher end of expected real-world gains.
Those are impressive numbers.
They’re also vendor-published results, not a promise about what will happen inside your acquisitions department.
And that’s where investors need to separate four things:
Fast is performance.
Cheap is economics.
Typed is interface reliability.
Correct is investment judgment.
Those are not the same thing.
Imagine you’ve told Jev that its only three possible answers are:
PURSUE
HOLD
PASS
And it returns:
PURSUE — 92% confidence
That feels definitive, doesn’t it?
But what exactly did the model prove?
It proved that it could return one of your three allowed answers in the correct format.
It did not prove that pursuing the deal is the right investment decision.
Maybe the OM is missing information.
Maybe your criteria were poorly defined.
Maybe the broker is using aggressive expense assumptions.
Maybe taxes will reset after acquisition.
Maybe your own buy box is outdated.
Maybe the model is simply wrong.
Even TypeSafe’s documentation notes that calibration across predictions doesn’t guarantee an individual answer is correct.
That’s why we keep coming back to this:
Confidence helps you route decisions. It doesn’t turn uncertainty into truth.
The Two Extreme Takes on Jev Are Both Missing Something
Whenever a new AI tool launches, it seems like we immediately split into two camps.
One side says:
“This changes everything.”
The other says:
“It’s just another AI model.”
We don’t think either is particularly useful here.
“Jev solves automation” is too broad.
“Jev is just another chatbot” misses what actually makes it different.
Where Jev appears genuinely interesting is high-volume, decision-shaped work where the choices are clearly defined, the criteria are specific, and code still owns the hard rules.
Which brings us to something we’ve been thinking about a lot lately.
Because there’s already an AI mistake happening inside real estate investment companies—and you don’t need Jev to make it.
The OM-in-Chat Mistake
You know the workflow.
A deal lands in your inbox.
You download the 40-page offering memorandum.
Upload it into ChatGPT.
And ask:
“Is this a good deal?”
A few seconds later, you have what looks like a mini investment memo.
Market overview.
Property highlights.
Risks.
Upside.
Maybe even a recommendation.
And for a moment, it feels like you’ve just compressed three hours of work into three minutes.
Except then somebody opens the rent roll.
The unit mix doesn’t reconcile.
Then someone checks the T-12.
The expense story is different.
Then you look at property taxes.
They’re going to reset.
And then someone finally asks the most embarrassing question of all:
Wait… did this deal even fit our buy box?
It didn’t.
You got a beautiful answer.
To the wrong question.
And this is where we think a lot of AI conversations are backwards.
The First Question Shouldn’t Be “Is This a Good Deal?”
Before we ask AI to write the memo, there are much simpler questions we need answered.
Does this deal fit our mandate?
Do we have the required documents?
Do the documents agree with one another?
Which assumptions came from the broker?
Which assumptions came from us?
Is there enough here to justify spending human underwriting time?
We call these glances and gates.
And they’re incredibly important.
Because only after a deal clears those gates should your team spend meaningful time underwriting it, researching the market, structuring financing, stress-testing assumptions, and preparing an investment memo.
Otherwise, something expensive starts happening.
We call it the dead-deal tax.
The Dead-Deal Tax Nobody Puts on the P&L
The dead-deal tax doesn’t show up as a line item in your financial statements.
But it’s everywhere.
It’s the analyst spending four hours underwriting a deal that never fit your criteria.
It’s the partner reviewing version seven of a spreadsheet.
It’s the Slack thread with 43 messages.
It’s the market research.
The broker calls.
The questions.
The revisions.
The attention.
All spent on a deal that should have been eliminated near the front door.
That’s the dead-deal tax.
And here’s the uncomfortable part:
AI can actually make it worse.
Because when AI produces a beautiful memo in seconds, it creates the feeling that progress has been made.
But readable information isn’t the same thing as relevant information.
Fluent language can make us feel like we’re moving forward before we’ve established whether the deal deserves our attention at all.
And that’s why the biggest lesson here isn’t really about Jev.
A Decision Tool Is Not a Deal Process
This is the sentence we’d put on a sticky note above every AI-powered acquisitions team’s desk:
A decision tool is not a deal process.
Because real estate deals don’t arrive as perfectly organized prompts.
They arrive as chaos.
An email.
A Dropbox link.
An OM.
A partial T-12.
A rent roll.
Three broker comments.
A revised rent roll two days later.
An insurance quote.
A tax estimate.
A market study.
And somewhere in the middle of all of that, assumptions start changing.
Then the deal moves through an entire lifecycle:
Screen → collect → underwrite → investigate → stress-test → decide → close → operate.
One AI response doesn’t give you that continuity.
It doesn’t matter whether that response is ChatGPT writing a three-page memo or Jev returning a beautifully typed score.
That’s why we’ve become much more interested in systems than individual AI tools.
What This Looks Like Inside Our Own Investment Process
This is the thinking behind AIREI OS.
We don’t think of it as a chat window with a really good memory.
We think of it as an operating system for the deal.
Because the deal itself needs somewhere to live.
The OM.
T-12.
Rent roll.
Notes.
Findings.
Underwriting.
Decisions.
When somebody asks six weeks later:
“Why did we pass on this?”
we shouldn’t need to search through five Slack threads and somebody’s Downloads folder to figure it out.
The history should live with the deal.
And then comes something even more important.
Your AI Can’t Follow a Mandate You Never Defined
Before AI evaluates a deal, we need to know what we’re looking for.
Markets.
Deal size.
Returns.
Structure.
Unit count.
Basis.
Hard rules.
Soft flags.
And blank information can’t quietly become a pass.
Because ChatGPT will happily help you analyze a beautiful 80-unit apartment deal even if your investment strategy says you only buy 150+ units.
Jev can evaluate whether something fits your criteria.
But only if you actually gave it criteria.
That’s the difference between asking AI to think for you and building AI into a process that reflects how you already think.
Then We Separate Facts From the Broker’s Story
Once a deal gets through the initial gate, we want the package converted into structured deal facts.
Then we underwrite using our math.
Not the broker’s story.
Not the OM’s perfect scenario.
Ours.
Hold versus refi.
Pro forma.
LP and GP economics.
Sensitivity.
Stress.
Chat can help explain those numbers.
Jev might eventually help gate certain decisions around them.
But neither should become the spreadsheet.
That’s a distinction we think is going to matter more and more as AI gets faster.
Let AI interact with your underwriting. Don’t let AI quietly become your underwriting.
The Truth Often Lives Between the Documents
This might be one of the most valuable things we’ve learned from evaluating real estate deals.
Sometimes the red flag isn’t sitting inside one document.
It’s hiding between two documents.
The OM says occupancy is one thing.
The rent roll says another.
Taxes look suspiciously low.
A property management fee seems to disappear.
Income doesn’t reconcile.
Each individual document might look perfectly reasonable.
The contradiction only appears when somebody compares them.
That’s exactly the kind of thing a good AI-enabled process should surface before you burn an entire investment committee cycle debating the deal.
But even then…
There is still one step we don’t outsource.
High Confidence Is a Signal. It Is Not a Signature.
Eventually, somebody has to make the call.
Should we move this deal forward?
Do we change our assumptions?
Do we call the broker?
Do we retrade?
Do we walk?
And when the answer isn’t obvious, we don’t want software pretending ambiguity disappeared just because a model produced a confidence score.
AI can help us interrogate the deal.
It can help surface conflicts.
It can prepare the investment committee.
But before capital moves?
A human still has to promote or approve the decision.
Because high confidence from a model is useful information.
It isn’t a signature.
So Where Does Jev Actually Fit?
This is where we think the conversation gets much more useful.
Instead of asking:
“Can Jev automate acquisitions?”
Ask:
“Which specific decisions should Jev make inside our acquisitions process?”
That’s a much better question.
Because different tools are good at different jobs.
Let software enforce deterministic rules:
Geography.
Price limits.
Unit counts.
Required files.
Calculations.
Let a decision model like Jev handle narrow semantic judgments:
Do these documents appear inconsistent?
Which risk bucket applies?
Does this package deserve further review?
Let chat do what chat is incredibly good at:
Explain.
Summarize.
Draft.
Help humans interrogate the information.
And then let the human own the question that ultimately matters:
Are we willing to put capital behind this?
That’s the division of labor we think makes sense.
The Most Important Question About AI Is Changing
For the last few years, everyone has been asking:
Which AI should we use?
ChatGPT?
Claude?
Gemini?
Now Jev?
And those questions matter.
But we’re starting to think there’s a more important question:
Where, exactly, should this AI earn a place in our process?
Because Jev might make certain glances faster, cheaper, and more consistent.
That’s valuable.
ChatGPT might make a complicated deal package dramatically easier to understand.
Also valuable.
But neither tool chooses your mandate.
Neither repairs missing data.
Neither owns your underwriting math.
Neither negotiates your LOI.
Neither sits across the table from your investors.
And neither has to live with the asset after closing.
That’s still us.
The Future of Real Estate Investing Isn’t AI Making Every Decision
We don’t think the future looks like a giant AI brain sitting in the middle of an investment company deciding what to buy.
We think it’s much more practical than that.
Machines handle the repeatable glances.
Code enforces the rules we already know.
AI helps us interpret ambiguity.
The deal maintains its history from first look to final decision.
And humans retain judgment where context and capital matter.
That’s a much less magical story than:
“Upload your OM and AI will tell you what to buy.”
But we think it’s a much more powerful one.
Because the goal isn’t fewer decisions.
The goal is to stop spending expensive human judgment on decisions that a machine, a rule, or a gate could have handled before the deal ever reached your desk.
So yes, we’re paying attention to Jev.
We’re excited about what faster, cheaper, structured decision models could unlock for real estate investment firms.
But the model isn’t the process.
And speed doesn’t replace judgment.
The real opportunity is building an investment process where AI handles what it does best—so humans have more capacity for the decisions only humans should be making.
Want to See How We’re Actually Building This?
If you want to see how we’re actually running AI systems inside a real estate investment firm—not theory, not another collection of prompts, but the actual stack and workflows—we’re breaking it down inside our free AI real estate investing masterclass.
We’ll show you how we’re using AI to increase the capacity of our investment team while keeping the investment judgment where it belongs.
Comments +