Will AI replace real estate agents? What the iBuying record shows (2026)

If you sell homes for a living, you have heard the forecast many times. Software will value the house, match the buyer and write the offer, and the commission goes away.

That forecast has already been tested with real money. Not in a demo, but on two public company balance sheets, with the results filed with the SEC.

The test was iBuying. A company makes an instant cash offer priced by its own valuation model, buys the house, and resells it. For the seller, the algorithm’s number stands in for the pricing advice an agent gives.

Zillow ran it on its Zestimate model. In February 2021, CNN recorded, Zillow announced that for certain homes “its so-called ‘Zestimate’ would also represent an initial cash offer” [source]. Nine months later it booked a write-down of approximately $304 million and began winding the business down [source].

What AI in real estate means

AI in real estate is the use of models that value, price, match or screen inside a property transaction, from automated valuation models to tenant-screening scores. The ones that matter are the ones that decide, because a decision has a price when it is wrong.

This post is about that narrower set. It is not a survey of listing-copy tools or chatbots, which save an agent time but do not take the call.

Will AI replace real estate agents?

No, not on the record so far. The two largest attempts to let an algorithm set the price on homes at scale both ended in large write-downs and layoffs, and the reason their own executives gave was the price forecast itself.

The obvious reading of Zillow is that the technology was early and will get there. That reading skips what Zillow said. Its CEO Rich Barton gave the reason in the company’s SEC-filed release: “the unpredictability in forecasting home prices far exceeds what we anticipated” [source].

That is not a complaint about a model being too slow. It is a statement that the thing being forecast moves in ways the model could not hold. An agent pricing a house carries the same uncertainty, but carries it one house at a time, with a seller who decides.

Zillow’s write-down was for “purchasing homes in Q3 at higher prices than the company’s current estimates of future selling prices” [source]. It warned of another $240 million to $265 million in Q4 losses, and cut its workforce by approximately 25% [source].

Will AI ever replace real estate agents if the market is calmer?

Opendoor is the answer to the “Zillow was a one-off” objection. It kept running the same model after Zillow left, and a year later the housing market turned and its model’s earlier purchases were marked down too.

Opendoor reported a net loss of $928 million for the third quarter of 2022, including “an inventory valuation adjustment of $573 million” [source]. Its own letter called the pricing engine an advantage: “we can adjust spreads rapidly based on changing market conditions” ([source]).

The spreads moved; the homes already bought did not. Weeks later Opendoor cut about 550 people, approximately 18% of the company [source].

Two companies, two years, the same failure: the model owned the inventory it priced, so every pricing error became a balance-sheet loss. An agent never owns the house, and that is not a weakness in the agent’s model. It is the design.

How will AI affect real estate when it decides who gets a home?

AI in real estate also decides who gets to rent, and there the record shows the liability landing on the company that sells the score. The question for an agent is the same one: who answers when the algorithm’s call is wrong.

Housing-voucher applicants sued tenant-screening company SafeRent over its SafeRent Score, alleging it discriminated by race and income [source]. Those were allegations, settled without any finding of discrimination.

Under the settlement, SafeRent may not put a score or an accept/decline recommendation on reports for voucher applicants for five years; the landlord gets the underlying information and decides [source]. The algorithm’s verdict was removed and a person’s judgment was put back.

The proof

TIN’s case file on Zillow checks every figure against Zillow’s own SEC-filed release and three independent newsrooms. It shows an AI valuation model turned into automated offers, then unwound at a loss: Zillow Offers and the $304M write-down.

TIN’s case file on Opendoor checks the loss and the write-down against two SEC filings and the layoff against three sources. It is the second test of the same model, a year later, with the same result: Opendoor’s Q3 2022 loss.

TIN’s case file on SafeRent reads the court-approved settlement and its non-admission clause. It is the rental-decision counterpart, not an agent-replacement case: Louis v. SafeRent.

Read the limits honestly. This is the iBuying record plus one screening case, not a survey of agent headcount. None of these files measures how many agents AI has displaced, and this post does not claim a number for it. On commercial real estate agents the corpus holds no case at all.

What each case actually settled

The bottom line

The forecast that AI replaces agents assumes the hard part of the job is the paperwork. The two largest public tests say the hard part is the price, and that a model which owns the consequences of its own price has to be right far more often than one that only advises. Zillow and Opendoor are not proof that agents are safe forever.

They are proof that the expensive part of the job is the judgment, and that a model taking that judgment must also take the loss.