Where's the AI?
Double-header this week because we launched and honestly… I just had to get this off my chest.
One of the biggest questions I find myself having every time I listen to Opendoor’s management or track press releases is simply...
“Where’s the AI?”
I think by now, most businesses understand that modern AI systems are the most important and impactful technology developed since… ever. The ability to mass-produce high-quality intelligence has far ranging consequences and upending the real estate industry is no exception.
To me, the most immediate application to Opendoor’s business is using AI to vastly improve pricing accuracy, segmentation, and, most importantly, market coverage.
In fact, back in 2023, Carrie already hinted at working on this in their Q2 2023 Earnings Call
“An example is our continued investment into home condition, which relies on computer vision, AI-based condition modeling and interior assessments, all of which give us more structured data to improve our overall data insights, which in turn informs home level pricing and pricing model accuracy.”
Now two years later - this is where we are.
“In Q1, we continue to add new features to our algorithms like school district quality and active competition.”
- Carrie during the most recent Q1 2025 Earnings Call
I’m no Carrie-hater by any stretch - it’s obvious that she's been dealt an extremely difficult set of cards - BUT I cringed when I heard this on the call.
11+ years of working on disrupting real estate, with access to some of the smartest PhD’s in the Valley, and looking at school district and active comps is what the company is touting in 2025?
Honestly, this sounds harsh, but these seem like the most basic features that any lay-person off the street would have thought to add to a V1 pricing algorithm.
In fact, there have been sadly no mentions at all of the AI-powered computer-vision features being worked on back in 2023 - despite huge improvements to both base foundational LLMs and vision models (see Tesla FSD / Optimus) over the last 2 years.
So what should be possible?
Back when I worked at Opendoor (and I’m guessing even today), we used essentially a human-computer hybrid to do home valuations. An internal human “appraiser-type” would use the base price from Opendoor’s valuation model and make tweaks based on “judgment" and "experience." For example: awkward, non-standard floor plans was a common trait that was hard to capture with any public data or quantitative features but that an human expert could spot and adjust pricing for.
Honestly, I think the company was on the right track back in 2023 with using AI computer vision to better understand home condition but actually wasn’t thinking anywhere big enough. There’s no reason why Opendoor shouldn’t be able to build an AI system that takes in or scrapes images, home walkthrough video, various seller commentary, public data (HOA, news, Nextdoor postings, etc.) and replicates the “valuation expert” at scale.
From my understanding, Opendoor already collects proprietary internal and external home pictures, home assessment reports, and now even video walkthroughs and commentary that can't be found anywhere else. It's hard for me to understand why they aren't testing what the latest AI models can do with this unique data, especially as vision and video models get more and more powerful.
The beauty of creating an army of "AI Home Valuation Agents" is that doing so will allow Opendoor to vastly expand its buybox (or homes it makes offers to) to those outside of just “cookie-cutter” sunbelt homes. One of the main reasons Opendoor doesn’t operate at scale in markets like Manhattan or the Bay Area is that the company isn’t confident that it can build an accurate enough pricing model to capture the “unique” characteristics of the many home types in those markets.
However, from a first-principles perspective, that logic’s never really held up in my mind. If a Manhattan-specialist real estate agent can walk through a 19th-century brownstone on the Upper West Side and quickly make an assessment based on knowing exactly what to look for - then it should definitely be possible with AI.
The system I’m imagining feels roughly analogous to how AI is currently being used in medical diagnostics - these models absolutely thrive when asked to pull in and process many disparate pieces and modalities of information and create a singular (or even range of) diagnostics / valuations.
Path to Medical Superintelligence: Doctors vs. AI Models in Diagnostic Accuracy
$OPEN AI
I understand the realities of building something like this are not easy by any stretch of the imagination, but even letting the public and investors know that Opendoor is working on something this ambitious (FSD anyone?) would be a huge shift in the narrative of what the company represents.
In fact, I imagine there’s a huge opportunity being missed RIGHT NOW to partner with PLTR, xAI, NVIDIA, OpenAI, Google, Microsoft, Meta, or any other leading-edge AI Lab to collaborate on Opendoor’s extremely unique housing-related data set and build absolutely the most accurate home pricing AI model in the world. In return, Opendoor can provide access to a source of data that I'm sure companies like Surge, Scale, Figure, Optimus would love to have in their training data (proprietary labeled and tagged housing reports, pictures, and videos)
If Opendoor started making the investment today; I'm picturing in 2-3 years their AI model could price a $3M New York Brownstone or $5M Palo Alto mid-Century Eichler as well as a $350K Phoenix starter-home stamped out by a builder in 1990. This would open the buybox to essentially 100% of the $2T total TAM, and just as importantly, help Opendoor rediscover its original “tech disrupter” roots.
Please subscribe if you liked this article!
I've been a die-hard Opendoor follower, employee, and shareholder since I joined the company in 2016. In many ways, the company has defined my professional and personal career. Currently, I work at DeepSky (founded by former Opendoor alum) where we're building the next business super-agent.
