The PracticeThe PracticeThe PracticeOperating infrastructure

The Practice · Guides

Will AI replace real estate agents?

Verified October 29, 2026 against each source at its publisher: the U.S. Bureau of Labor Statistics, Science, the International Labour Organization and NAR.

The short answer

The Practice's position is no: AI replaces parts of the job, not the job. The practice stays, meaning the judgment about a price or a counter, the relationships that bring the next client, and the accountability that comes with a license.

That is a view argued from the evidence below, not a forecast. The strongest case against it is stated in full further down.

What do the federal projections say?

The U.S. Bureau of Labor Statistics projects employment of real estate brokers and sales agents to grow 2 percent from 2025 to 2035, from 530,600 jobs to 538,800. That is slower than the average for all occupations, and it is growth, not decline.

The same projection expects about 40,400 openings a year on average, most of them from people who change occupations or leave the labor force. The Bureau ties the outlook to housing demand, credit, prices and the economy. A projection is not a promise, and the next edition may move.

What does the research on AI and tasks say?

The research measures tasks, not jobs. A study published in Science in June 2024 by Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock estimates that roughly 1.8 percent of jobs could have over half their tasks affected by large language models with simple interfaces, and just over 46 percent once software built on top of the models is counted.

The International Labour Organization's 2025 index reaches a similar shape from another direction. It finds one in four jobs worldwide potentially exposed to generative AI, and calls "transformation, not replacement" the most likely outcome, because many tasks done faster still need a person.

Which parts of the job does AI take?

AI takes the parts made of reading, writing and organizing information. Those are what large language models do, and an agent's week holds plenty of them.

Of the ten duties the Bureau lists for the occupation, several begin as text: lists of properties with their details, advertisements and listing copy, the comparison behind a price, and the paperwork that follows. In a working week that means the listing description, the first cut of a market comparison, the follow-up email, the summary of a long thread, and the checklist of dates.

An assistant can draft every one of those. None of them is the decision.

What stays with the agent?

Judgment, relationships and accountability stay. They are also what buyers and sellers say they hire an agent for.

In NAR's 2025 Profile of Home Buyers and Sellers, 88 percent of buyers and 91 percent of sellers used a real estate agent or broker. NAR reports that buyers want an agent who can help them negotiate and understand the market, not only find the home.

Judgment: which comparable counts, when to counter, when to tell a seller the price is wrong. An assistant can rank the options; the agent decides which one the client hears.

Relationships: repeat and referral business comes from people who remember a person. An assistant can schedule the call; the client answers the agent.

Accountability: the license, the supervising broker, and a name on the advice. No vendor signs the advice an agent gives.

What is the strongest case for replacement?

The strongest case says the job is mostly information work, and information work is what these models do. It is a serious argument, and it deserves to be stated at full strength.

It points out that buyers already search listings on their own, and that the Science study's 46 percent figure counts software that is still arriving. It adds that an occupation with about 40,400 openings a year can shrink by hiring fewer people rather than by letting anyone go, so the employment total could hold while each agent's share of the work changes underneath it.

How does The Practice answer it?

The case is right about tasks and wrong about the job. Exposure measures what a model could speed up, not what a client will hand to a model, and the Science study scores tasks affected, not jobs removed.

Three things a model does not supply: a license and a supervising broker; a person a nervous seller will call at nine at night; and a record in one market that a buyer can check. The Practice's view, not a forecast: the work shifts toward fewer hours on drafts and lists and more on the conversations only the agent can have, and the gap widens between agents who build that system and agents who do not.

What should an agent do about it?

Hand the information work to an assistant and keep the practice for yourself. Start with the drafts, lists and summaries named above, keep a person on every judgment and every client conversation, and read every draft before it leaves.

The free guide Your first weekend: seven setups sets up the account in about two hours.

Next rung

The Letters carry this argument forward twice a month, by email. Each one says what changed in AI for a working practice, sourced and dated.

Revision history

1.0, October 29, 2026, first version. Figures opened at each publisher on October 29, 2026.

Educational, not professional advice. Nothing on this page is affiliated with or endorsed by its publisher.