I recently listened to a conversation between Ethan Mollick, Associate Professor of Management and Faculty Director of the Generative AI Labs at the Wharton School of the University of Pennsylvania, and Simon Sinek on Sinek's A Bit of Optimism podcast. They discussed something I've been thinking about quite a bit lately.
If everyone has access to powerful artificial intelligence, what makes one professional more valuable than another?
AI can now produce an impressive market report, build a website, develop a marketing strategy or analyze substantial amounts of data. Mollick suggested that as AI raises the quality of everyone's work, human experience, perspective and judgment may become increasingly important. A recent assignment helping a real estate colleague illustrated that point for me.
A Building I'd Never Heard Of
A colleague recently asked me to help research an oceanfront condominium building in South Florida. She was considering pursuing a listing there and wanted to understand the building and its market. I had never heard of it. That wasn't unusual: South Florida has hundreds of condominium buildings, each with different sales histories, ownership structures and competitive positions.
But I did know how to approach the research. We started with MLS history, property tax records, current listings and comparable condominium transactions. Working with my AI research assistant, we developed what became a 20-page listing brief. The AI organized records, calculated statistics, produced charts and assembled research. My job was to decide which questions mattered and which conclusions the evidence could actually support.
When the Numbers Tell the Wrong Story
One residence sold in 2020 for $2.3 million. Another sold in 2024 for $3.5 million. Both were reported at 3,666 living square feet. That's a 52% difference in sale price, a tempting figure to present as appreciation.
But these were different residences. Were their views comparable? Had one been extensively renovated? Were their floor levels, outdoor spaces and finishes equivalent? Until we investigate those differences, the numbers tell us what two properties sold for, not how much a particular property appreciated.
The sample was also thin: the supplied MLS history contained six closings from 2020 onward, with several years showing no closings in that particular export. A polished chart could easily suggest more precision than the data justified. The percentage calculation took seconds. Understanding what it meant required considerably more thought.
A Small Data Error With a Bigger Lesson
An older MLS record classified one unit as a townhouse. A strict condominium-code filter would have excluded it. The tax records, however, confirmed that it belonged to the condominium building. Reconciling the sources changed our historical expired-listing count from 17 to 18.
Those 18 expired listing events represented only nine distinct residences. Some units had been marketed unsuccessfully more than once. Dividing expired events by closed transactions to calculate a failure rate would have mixed different units of analysis. A listing is a marketing event; a condominium unit is a property. One unit can be listed repeatedly and eventually sell.
We also couldn't establish a reliable cancellation count because the necessary status information wasn't present in the supplied data. We said so. That is a useful result in itself. Domain expertise helps you recognize when the data is answering a different question from the one you intended to ask.
Knowing What to Compare
For larger residences, we examined nearby closed condominium sales between $3 million and $5 million. In our selected comparison group, sold prices ranged from roughly $790 to $1,719 per living square foot.
That's an enormous spread for properties within a relatively narrow purchase-price band. But oceanfront buyers understand that floor level, views, renovations, building services, assessments and residence configurations can make two similarly priced condominiums very different purchases.
The AI identified transactions and organized comparisons. I defined the selection criteria and the questions that needed further investigation. The brief gave my colleague a stronger starting point for a conversation with an owner, while identifying what she still needed to verify before recommending a listing price.
Experience Changes How We Use AI
In his conversation with Simon Sinek on A Bit of Optimism, Ethan Mollick discussed research involving Boston Consulting Group in which junior professionals sometimes performed worse with AI despite appearing comfortable using it. Their ability to operate the technology wasn't necessarily the problem. Evaluating the output required knowledge they had not yet developed.
I've spent approximately 18 years helping real estate agents with marketing, technology, pricing strategy and business development, including preparing for listing presentations. That experience shapes what I ask AI to do. I generally have a sense of what the records should contain, what comparisons are appropriate and what might go wrong.
Experience doesn't make anyone infallible. It can create blind spots, and newer professionals often bring valuable fresh perspectives. But familiarity with an AI interface and understanding the underlying business are two different capabilities.
The Value of Evaluating the Work
Sinek observed that as AI becomes faster at complicated tasks, efficiency increasingly depends on how quickly we can determine whether the work is correct. Mollick connected that observation to expertise.
Our condominium project illustrates the point. AI assembled much of the research quickly, but the building identity, transaction classifications, comparison choices and statistical interpretations still required scrutiny. Some issues were easy to spot. Others took additional research. Some questions remained unanswered.
The benefit was that I could spend more time on the consequential decisions rather than manually assembling every table. That is a more useful way to think about AI productivity than counting how many pages it can produce.
Teaching Agents to Challenge AI
Early AI education focused on prompting. That remains useful, but modern systems are increasingly good at interpreting ordinary instructions. I believe the next phase of training should put more emphasis on evaluation.
Mollick recommends asking AI to examine work from different perspectives: first as someone unfamiliar with the topic, then as a skeptical expert searching for weaknesses. For a pricing analysis, I might ask which conclusions rely on thin samples, which property differences have been overlooked, what another experienced agent would challenge and what new evidence might change the recommendation.
Those questions improve the process, although the AI's answers still require independent verification. I'd rather work with an agent who can challenge a market analysis than one who can generate an impressive report without understanding it.
The Bigger Opportunity
My colleague asked for help understanding a building. Together with AI, we developed a detailed research package covering sales history, listing attempts, competing properties and questions for a potential seller. Preparing a similar package manually would have required substantially more time.
Now imagine extending that capability across dozens of communities, hundreds of condominium buildings and thousands of properties. That's part of what I'm building through the market intelligence resources on DavidAbernathy.com: making deeper property research accessible while retaining the analytical discipline needed to use it responsibly.
AI will keep improving. Producing a polished report or calculating market statistics may soon be routine. Someone still needs to decide what deserves investigation, which information can be trusted and what the findings mean for a client.
AI gives us access to extraordinary analytical capabilities. Domain expertise helps us know what to do with them.
Case study: anonymized South Florida condominium research using supplied MLS and property tax records through October 8, 2026. Transaction comparisons are unadjusted and do not establish property-specific appreciation or valuation. The original research brief documents the source limitations.
Related: Why I'm Building an AI Brain · AI and Public Real Estate Data · More market intelligence.