My AI and Your AI, Sitting By the Fire

Real Estate Technology
1moFollow

My AI and Your AI, Sitting By the Fire

Steve K.
Realtor · Boulder, CO · Member since 2016 · 3k+ posts · 5k+ votes

There has been some good discussion on here regarding the role of AI in the real estate market. So I thought I’d share my recent first-hand experience with it. Both buyer and seller were using AI heavily to make their decisions. Seller asked AI about list price, how to prepare the property etc. All things selling were run through AI before making any decisions. The buyer did the same. Every piece of advice they received from humans was scrutinized heavily through their AI. The deal went about as you would expect it would. Rather horribly for everyone involved. Here’s one example: during inspection objections, the buyers AI somehow came up with the number of $60-80k that they should get back in seller concessions for various miscellaneous inspection items despite having very little ammunition to negotiate with in reality (the house was in great condition, the appraisal came back $20k over asking price, and the seller had a cash, as-is backup offer that was higher than the contract amount). Basically, the buyers AI was on crack. The sellers AI of course recommended taking the higher backup offer. Needless to say, the (human) agents on both sides had to work overtime to bring some logic back into focus. AI can be helpful as a super-powered internet search, but ultimately there is a human on the other side of the deal. These days, a human and their AI, apparently. Buyers and sellers still need to use their brain, basic logic, and empathy. We ended up making it to closing, somehow.

1Reply
360 views

3 Replies

Jump to latestLatest
  • Michael Eskenasy
    Investor · Pacific Northwest · Member since 2026 · 538 posts · 307 votes
    1mo

    This is actually a good example of why “using AI” and having an intelligent decision system are not the same thing.

    The buyer’s AI didn’t magically become irrational. It was almost certainly reasoning from an incomplete representation of the deal. If someone feeds it an inspection report and asks, “What concessions should I request?”, it can generate a theoretically defensible number based on repair costs. But unless the model also knows the appraisal came in $20K high, the property condition relative to comps, the seller’s leverage, the existence and strength of a higher cash/as-is backup, the actual contract language, local norms, and which inspection findings are genuinely material, it has no idea what the buyer’s negotiating leverage actually is.

    So $60–80K can be mathematically explainable and strategically absurd at the same time.

    Same problem on the seller side. “Take the higher backup offer” sounds obvious until you ask whether the current contract can legally be terminated, what the contingencies say, what litigation/default risk exists, whether the backup is actually executable, and what probability you assign to each closing. AI that doesn’t maintain the whole transaction state is just giving advice from whatever slice of reality somebody happened to type into the box.

    That’s the failure mode I think people are missing: AI compresses the information you give it; it does not automatically possess the information you forgot to give it.

    The answer isn’t “AI bad, humans good.” Humans make exactly the same mistake when they give advice without enough context. The difference is that AI can make incomplete reasoning sound extremely complete.

    Used correctly, I’d want the system tracking the evidence, contract state, appraisal, inspection issues, alternatives, leverage, counterparties, assumptions and unresolved questions continuously — and then letting the humans make the judgment call.

    What you experienced wasn’t really too much intelligence in the transaction. It was two very confident machines working from partial files.

    1Reply
  • Curtis Sellon
    Investor · Cumberland, RI · Member since 2025 · 1 post · 1 vote
    1mo

    Michael is spot on, and I think we’ll be seeing much more of this over the next year or two.

    Currently, AI works primarily from the data we feed it. If critical information is missing (whether intentionally or simply because the person didn’t realize it) that becomes its weak point.

    As AI advances and begins pulling from more sources (email, text messages, documents, even those smart Ray-Bans during a property walkthrough) it’s going to miss a lot.

    I’d still tell clients early that they can and should use AI. They’re probably going to use it anyway. But remind them that its recommendations will be heavily weighted/bias toward the facts and context they gave it.

    If they’re using ChatGPT, I’d also recommend setting up a Project for the transaction rather than relying on a single chat. Upload the files there and keep the transaction information together so the AI maintains more consistent context as the deal progresses.

    And this probably goes without saying, but I’m not a SecOps guy: feeding confidential, financial, contractual, or client information into a third-party AI platform can have privacy, retention, and compliance implications that people need to understand.

    AI is becoming very good at reasoning over the information it has. The trick is remembering that it has no idea what you forgot to tell it.

    1Reply
  • Steve K.
    OP
    Realtor · Boulder, CO · Member since 2016 · 3k+ posts · 5k+ votes
    1mo

    Yes the problem is that AI is limited to what it can learn from the internet, which can be both helpful at times but also full of errors and is never the complete picture in real life. In this case the inspection objection was full of mistakes and that worked against the buyer for example it pulled pricing for repairs that were not accurate and used the assumption that the house should be in perfect condition (no such house exists) and didn’t factor in the seller’s perspective/ had no empathy. This damaged the buyers credibility with the seller/ listing agent and wasted everyone’s time unnecessarily. AI is not only often just plain wrong, it also can’t apply logical thinking, look at the big picture or greater context, or use common sense. It will never be fully aware of all that goes on during a transaction between humans in the real world, or master the artful nuances of negotiating effectively towards a win-win in good faith. Plus AI has strong confirmation bias, each side’s AI is going to tell each side what they want to hear based on their inputs. I can easily get AI to confidently agree that the word Seventeen only has one E, but that doesn’t make it true. It takes both willing seller and willing buyer to make a deal happen. This deal would have definitely blown up without the two human agents bringing both sides back to reality.

    0Reply
Join the conversationCreate a free account to reply, vote on answers and follow this thread.
Log InJoin for Free