The honest answer is no, and the reason has nothing to do with the human touch, relationships, or any of the other comforting phrases that fill the first page of results for this question.
The answer is no because the federal agency that projects American employment for a living has already run the numbers, and because a version of this exact story played out thirty years ago in another industry and we know how it ended. Both are worth looking at, because the real threat to your income is sitting somewhere else entirely, and almost nobody writing about this points at it.
What the official projection says
The Bureau of Labor Statistics publishes a ten-year outlook for every occupation it tracks. For real estate brokers and sales agents, the current numbers are these.
| Measure | Figure |
|---|---|
| Employment, 2025 | 530,600 (427,000 agents, 103,500 brokers) |
| Projected change, 2025–35 | +8,300 jobs, or +2% |
| Versus all occupations | Slower than average |
| Median pay, agents (May 2025) | $52,830 |
| Median pay, brokers (May 2025) | $73,220 |
Growth, not collapse. Slow growth, but growth. And here is the part worth sitting with: when the BLS lists what will actually move this number over the next decade, it names demand from people relocating and buying, sensitivity to the wider economy, tighter credit, rising prices pushing people into renting, and retirements creating openings.
Artificial intelligence does not appear on that list. Not as a risk, not as a factor, not at all. The people whose entire job is forecasting which occupations technology will hollow out did not consider it worth mentioning for this one.
The industry that already ran this experiment
Every conversation about this eventually reaches travel agents. The internet was going to erase them: booking sites did the search, the comparison and the transaction, and the middle person had nothing left to sell.
Thirty years on, the BLS counts 61,500 travel agents in the United States, with employment projected to change by 0% through 2035 and median pay of $50,160. The profession got much smaller and it stopped being a place for order-takers. It did not disappear.
What survived was the part of the job that was never really about access to information: the complicated trip, the group booking that falls apart, the client who wants someone accountable when the connection is missed. What died was the part that consisted of looking something up on behalf of a person who could now look it up themselves.
That distinction is the whole answer, and it maps onto real estate almost line for line.
What AI is actually taking
Not the job. The tasks that used to fill the job. There is a difference, and it decides who has a career in five years.
| Task | Status |
|---|---|
| Writing listing descriptions | Largely gone. A model does it in seconds, and does it well |
| Drafting follow-up and nurture emails | Largely gone |
| Producing market summaries and reports | Going. The writing is automated; the numbers still are not |
| Answering routine buyer questions | Going, and buyers are already asking a chatbot before they ask you |
| Selecting and adjusting comparables | Not close. Requires MLS access and local judgment |
| Pricing a specific house | Not close, for the same reasons |
| Negotiating a repair credit at 9pm | Not close, and not the kind of problem that gets solved by more data |
| Being accountable when it goes wrong | Cannot be automated. Somebody has to hold a license |
Read the top half of that table and read the bottom half. The top half is the work a new agent used to do to be useful in their first two years. The bottom half is what experienced agents do.
AI is not removing the profession. It is removing the on-ramp. That is a real problem, and it is a completely different problem from the one everybody is arguing about. Where exactly the machine stops being useful on the pricing side is worked through in the piece on what AI can and cannot do with a CMA.
The number that should actually worry you
Here is where the real pressure is, and it has been building since well before any of this.
As of June 2026, the National Association of REALTORS® reported 1,439,163 members, down from 1,463,352 a year earlier, against a peak of roughly 1.5 million in 2022. NAR’s own framing is that membership is holding up: staying above 1.4 million has happened only a handful of times in the association’s history. And it is holding up while existing-home sales run at their slowest pace since 1995.
Sit those two facts next to each other. Close to a million and a half licensed members. The fewest transactions in three decades. That is the squeeze, and no chatbot caused it.
The agent who loses their income over the next few years will not be replaced by software. They will be outcompeted by another human being for a deal that a shrinking market only had one of. The jobs software is genuinely taking over are narrower than the headlines suggest, and the ones it is creating for criminals — impersonating a seller convincingly — get less attention than they deserve. AI matters here, but only as an amplifier: it makes a good agent noticeably faster and it makes a mediocre one indistinguishable from every other mediocre one, because they are all now sending the same competent, generic, machine-written email.
The change that is real, and nobody frames it right
Forget replacement for a second. The thing that has already changed, quietly, is who does the first hour of thinking.
A seller used to call three agents and ask what their house was worth, because that was the only way to find out. Now they ask a chatbot first, at eleven at night, and they arrive at your appointment with a number, a rough plan and a list of things they have decided are true. Some of it is right. Some of it is confidently wrong in ways that are hard to unpick without sounding defensive.
That is not a smaller job. It is a harder one. Explaining why a plausible answer is wrong takes more skill than supplying the answer did, and it has to be done without making the person feel foolish for having asked.
The agents who struggle with this are the ones who treat it as a threat and get prickly. The ones who do well treat the client’s research as the opening of the conversation rather than a challenge to their authority: good, you have already looked at this, let me show you the two things that number cannot see. A model has no access to the MLS, no idea what the house two streets over actually closed at after concessions, and no memory of the inspection that killed the last deal on that block. Those are still yours, and they are still the whole game. Why a general assistant cannot reach that data at all is covered in the piece on ChatGPT and MLS access.
Handled that way, a client who arrives pre-researched is easier to work with than one who arrives blank. They already care, they have already spent an evening on it, and they are asking better questions than they would have three years ago.
What this means for what you do on Monday
- Automate the top half of the table without guilt. Every hour you spend hand-writing a listing description is an hour not spent on the work that cannot be copied. The prompts for MLS remarks and the ones for follow-up emails exist for exactly this.
- Get deliberately better at the bottom half. Pricing, negotiation, and the conversations nobody enjoys. Those are now the entire job, and there is no shortcut into them.
- Stop sounding like the machine. When every agent has the same tool, the generic version of anything is worthless. What survives is specific: this street, this buyer pool, this house’s problem.
- Assume your client already asked a chatbot. They arrive with an answer. Sometimes it is wrong, and correcting it politely and accurately is now part of the listing appointment.
- Watch what you say and sign. Speed makes it easy to send something you have not read. Fair housing obligations and the line around legal advice do not move because a model wrote the sentence, which is the subject of the piece on AI and contract questions.
Common questions
Could an AI legally represent a buyer or seller?
Not as things stand. Representation requires a license, licenses are held by people, and every state ties fiduciary duty and liability to a named individual. A regulator would have to decide who gets sued when it goes wrong. Nothing suggests any state is moving that direction.
What about the tools that promise to run the whole transaction?
They automate coordination, which is genuinely a large share of the work, and they are getting good at it. They do not remove the person whose license the file sits under. What they change is how many transactions one agent can carry, which pushes the same way as everything else here: fewer people doing more deals each.
Is it worth getting licensed right now?
Harder than it was, and not because of AI. The market has the fewest sales since 1995 and close to 1.44 million REALTORS® competing for them, and the entry-level tasks that used to let a new agent be useful are the ones being automated first. If you go in, go in knowing you need to be good at the difficult half of the job much sooner than agents did ten years ago.
Should I tell clients I use AI?
Usually yes, once, framed as competence rather than confession. When and how is worked through in the piece on telling a client you used AI.
If you want the practical inventory rather than the forecast, the overview of how agents actually use AI covers what these tools are good at and the four things they still get badly wrong.
Sources
- US Bureau of Labor Statistics, Occupational Outlook Handbook: real estate brokers and sales agents. Employment 2025, projections 2025–35, median pay May 2025.
- US Bureau of Labor Statistics, Occupational Outlook Handbook: travel agents. Employment 2025, projections 2025–35.
- National Association of REALTORS®, membership figures as of 18 June 2026 and context on the pace of existing-home sales.
Checked August 2026. Employment projections are estimates, not promises, and the BLS revises them.