{{first_name}},

My mom's cabin in Lakeside, Arizona has no foundation.

We bought it years ago for $40,000.

Here’s how the cabin was built:

Yes, the wheels were still on the trailer 😄.

The trailer became some of the square footage…which was RECORDED on tax records.

The connection between trailer and floor was janky on its best day.

And none of that scared us.

At $40,000, the land alone was worth more than the price.

Whatever was wrong above the dirt, the dirt covered it.

Then came the second number...

We’ve put over $100,000 into reverse engineering and rebuilding 800 sq ft. We had to open it to know how the “house” was even built.

$40,000 to buy it. Over $100,000 to find out what was holding it up.

{{first_name}}, that ratio is the most honest description of AI I can give you right now.

Cheap to acquire.

Expensive to trust.

Everyone budgets for the $40,000.

Almost nobody budgets for the $100,000.

This week I wrote out the eight problems I keep running into, building my own tools with AI and watching operators wire it into theirs.

The first seven explain why the verification bill always shows up after the thing is load-bearing.

The eighth is the reason I wrote the piece...

because it breaks the comfortable math.

With the cabin, our downside was poorly built flooring, inaccurate tax records, and a janky trailer we had to do something with.

With AI, there is no dirt…no safety net…no fallback.

1. It looks finished before it is finished.
The cabin passed the eye test: walls, floors, a roof. You had to open things up to find the trailer. AI output works the same way: clean formatting, confident tone, complete paragraphs. Polish is not structure. Polish is what hides structure.

2. It says wrong things in the same voice as right things.
A junior analyst hedges when they're unsure. You can hear it across the room. A model doesn't have an unsure voice. The invented number arrives wearing the same suit as the real one, and nothing in the delivery tells you which is which…you have to investigate, and AI is teaching us to trust and not verify.

3. The cost structure is inverted.
Generating is nearly free. Verifying is not. Everyone budgets for the $40,000 and almost nobody budgets for the $100,000, and the verification bill shows up at the worst possible time: mid-raise, mid-close, mid-audit. Yes, I saw a financial model that made no sense in a deck and knew immediately it was bunk…

4. The failures live at the connection points.
The cabin's real problem wasn't the trailer or the new flooring. It was the janky connection between them. Same with AI. One answer in isolation usually holds. The danger is the handoff: the output of one step feeding the next, tools wired to tools, until nobody can say where the load path actually runs.

5. It scales your mistakes at the same speed it scales your work.
A bad assumption in a document you wrote by hand stays in one document. A bad assumption in an automated workflow propagates into everything that workflow touches. The cabin was one house. A pipeline is a hundred of them, all built the same way. Imagine the missed details at scale…lives would be ruined.

6. Errors arrive without an author.
Somebody built that cabin. A person made each of those choices, and the work carries their fingerprints. When a model produces the work, accountability goes soft. "The AI got it wrong" is now a sentence people say out loud in real businesses, as if it settles something. An error with no author never gets owned, which means it never gets fixed.

Someone is ALWAYS behind an AI iteration.

7. The discount does your rationalizing for you.
We told ourselves it wouldn't matter if the cabin was messed up, because the price was that good. And for us, it was true. The same sentence is getting said about AI all over the industry: it's basically free, so who cares if it's a little wrong. Look at what's carrying that sentence. Free is doing all the work.

8. There is no land under it.
This is the one that breaks the analogy, and it's the reason the analogy matters. Your downside on the cabin was floored by dirt. Worst case, scrape it and you still own a lot in Lakeside.

There is NO DIRT under an investor letter with a wrong number in it. No salvage value in a model-built underwriting error that already priced the deal. When this foundation fails, you don't fall to the land value. You just fall.

The arc:

Yes, we are happy with our purchase of my mom's cabin up in Lakeside, Arizona.Yes, there was a very large margin of error available to us, which is why we actually bought it.And funny enough, that margin of error was tested. And it still made my stomach turn.Like the sloppy craftsmanship of what seems to be a great build, AI only requires one sophisticated set of eyes to read through the bull crap…then you’re toast…credibility is weakened.

Have a great weekend.

P.S. If you want to talk through what your product, reporting, or investor relations system needs to look like to be in the consideration set when this capital starts flowing, grab 15 minutes with me here: andrewlebaron.com/meetwithandrew

P.S.S. If you want a newsletter like this one to help you raise capital and stay relevant, I’ll help you build it…and no, not with AI. Book a call and let’s chat.

F6 Partners is an alternative real estate asset manager that I’m currently advising, specifically assisting in the buildout of their Investor Relations department to raise another $100M for Student Housing, and so far they are on a roll.

Learn more about F6 Partners →

This message, including any hypothetical scenarios described, is provided for informational and illustrative purposes only and does not constitute professional advice. These scenarios are hypothetical and are not indicative of any specific outcome or past performance. Results will vary based on individual efforts and external factors. We make no promises or guarantees regarding your success or income level.

Please note that individual successes are influenced by personal abilities, market conditions, and other external factors. We assume no responsibility for decisions made or actions taken based on the content of this email. Always consult with qualified professionals before making significant business decisions.