My prediction: will AI replace real estate agents in 2026?
In January 2026, a Delta Media survey covered by Inman found that 97% of brokerage leaders report their agents actively using AI.
A few weeks later, Ascendix put the daily-use figure at 87% across brokerages and agents. Set those numbers next to Gartner's forecast that 40% of enterprise applications will include task-specific AI agents by end of 2026 — up from under 5% at the start of 2025 — and you have the outline of a transformation that's already happened without most real estate agents realizing it.
My read is that the adoption curve is running ahead of the capability curve.
The AI that agents have integrated into their weekly workflow is still mostly the 2024 generation — generative models that draft and summarize.
The 2026 generation is different. AI is already causing mass disruption to every industry.
So we have to ask what makes real estate any different?
The tools agents are running day-to-day aren't impressive on their own and, if you’ve used any AI tools, you’ll know that’s true.
Each one shaves time off a specific weekly task, but none of them change what an agent fundamentally does.
Jason Ivens at KW Westfield in Orem, Utah runs 300 agents and told Keller Williams that his top 101 closed more deals and hit $183,000 median income after he pushed AI adoption two years ago.
That's a good headline until you read it carefully.
300 agents, only 101 hitting the top bracket, which means the bottom two-thirds either aren't using the tools at all or aren't using them well, and they're losing ground inside their own office to the agents sitting two desks away.
The AI-curious agents — the ones who tell their broker they're "experimenting" with ChatGPT for listing descriptions — lose market share month over month to the AI-native agents in the same office, who've built the AI tools into their pipeline.
The AI doesn't make the AI-native agents smarter or more skilled. It gives them back time. Ten extra hours a week compounded over 12 months is roughly 20% more client-facing capacity, and in a zero-sum market where every listing goes somewhere, that's where the transactions migrate.
By the time the AI-curious agents realise they're being outcompeted by their own colleagues, the gap is too wide to close without a full workflow rebuild most of them aren't going to do.
The part that gets missed in the "97% of brokerages are using AI" headlines is a simple distinction: using AI and benefiting from AI are doing very different jobs in these stats.
Most agents have the tools switched on and are barely touching them. A handful have built them into their workflow and are taking market share from the rest. McKinsey's estimate that AI could generate $110 to $180 billion in annual value for the US real estate sector is real, and I think it's probably low, but that value is heavily concentrated in the top 20% of agents who've figured out how to compound AI gains across a book of business.
The middle 60% get marginal gains, mostly in admin and content. The bottom 20% are handing their leads to the top 20% through slower follow-ups and weaker positioning. Call it a redistribution event. The transactions are getting closed either way.
Fewer agents are closing them.
IDC expects AI copilots embedded in roughly 80% of workplace applications by the end of this year, and PwC's Emerging Trends in Real Estate 2026 draws a line between today's generative AI and what they call agentic AI.
KW Command relaunched in February 2026 with direct API integrations to Gemini, RemyAI, and Rejig.ai, which means a KW agent's CRM now talks to three different reasoning models depending on the task, and that capability is sitting in the platform whether the agent knows how to use it or not.
My bet on what end-of-2026 delivers in real estate specifically: a workflow where inbound leads get qualified, scored, nurtured, and scheduled into showings without an agent touching the top of the funnel.
The agent gets involved at the showing and stays involved through the offer.
Everything before, between, and after — including post-close nurture and sphere-of-influence drip — gets absorbed by the stack. I'd put that at 70% probability by December 2026 for teams at Compass, eXp, Real, and Keller Williams, because they have the platform infrastructure and the budget to deploy it.
For independent agents without a tech-forward brokerage behind them, I'd put the probability closer to 40%, and dropping fast as the cost gap between a platform agent's stack and an independent's stack widens.
The tools that will matter most by December are those coming from the major models: Anthropic and OpenAI. Claude Cowork, as an example, can handle 80% of real estate admin tasks without much (or any) human interaction.
The model capability is already there. And it’s advancing incredibly quickly. Quicker than Sam Altman predicted. Quicker than any major AI player could’ve foreseen.
What the models won't be able to do by the end of 2026 — and probably not by the end of 2028 either — is sit across from a seller who's just been told their property is 15% overpriced and hold that conversation.
Walk a first-time buyer through their second round of inspection findings.
Read a room.
Tell someone what they don't want to hear.
The agreeable-by-default problem in current LLMs is baked in through alignment training. It's a feature, and the frontier labs have no incentive to change it (ask anyone who's tried to get ChatGPT to push back on a decision they've already made).
Negotiation and judgment survive. Everything else doesn't.
In August 2024, when the NAR settlement took effect, John Campbell at Stephens predicted a 50% agent attrition rate within two years, and the entire industry braced for it.
It didn't happen.
NAR's own mid-2025 numbers projected membership falling to 1.2 million by end of 2026 — down from a 1.6 million peak, a 25% drop — which is real, but slower than the doom forecasts.
Average buyer-side commissions, which were supposed to collapse under settlement pressure, actually held: Clever Real Estate's February 2026 data has them at 2.82%, higher than the 2.55% they averaged in early 2025, right after the settlement took effect.
The lawsuit didn't deliver what the plaintiffs wanted, and most of the analysts covering the space have spent the last 18 months explaining why.
I was one of the people saying AI wasn't coming for agents in any serious way this year. I read the tools right and the deployment curve wrong. Even AI SEO has been adopted far more aggressively than we could've ever anticipated.
What I thought was a 2027 problem is already here for most real estate agents.
The reason the settlement failed to move commissions is clients kept asking for human representation and sellers kept paying for it, and regulation doesn't override what buyers and sellers are willing to do in a market.
What moves markets is infrastructure. The commission settlement was a legal framework agents could resist by keeping their workflow unchanged, which is what happened.
AI is an infrastructure change the agents themselves are driving by upgrading their own workflow, which is why it's going to move the number the settlement couldn't.
If your job is the tasks — listing copy, follow-up emails, MLS admin, CMA prep, scheduled calls, drip campaigns, sphere-of-influence nurture — you're already being replaced, and the only reason you still have a job is that your broker hasn't fully deployed the stack and your clients haven't noticed the difference.
Both of those gaps close within the next 12 months. Max.
The software is cheaper, more consistent, and by December 2026 it'll be orchestrating the whole task pipeline without a human pressing go between steps. You're not competing with other agents anymore.
You're competing with a workflow that doesn't need to eat or sleep.
If your job is the decisions:
End-of-2026 AI won't touch you.
End-of-2028 AI probably won't either but that could change. The alignment problem means models will keep agreeing with whoever's typing into them, and the judgment tier requires the opposite: someone willing to tell a seller their expectations are wrong, or tell a buyer their dream house is overpriced by six figures.
The problem is most agents don't know which category they're in.
They think they're decision-tier because they've been doing this for 12 years and closed $8M in 2024. But if they audit their week, 90% of the hours were task-layer work.
The $8M closed on top of the tasks. Strip the task layer out and the closings stay. The software runs the pipeline.
The agent sees the offer on the screen where their CRM used to send it.
The best agents at the task layer are the first ones to go.
By December 2026, I think NAR hits its 1.2 million projection, which is 400,000 fewer agents than the peak. Most of them will blame the market. Some will blame commission compression.
A few will blame AI.
Almost none of them will connect it back to the choice they made six years ago, when they defined their career as a series of tasks instead of a series of decisions.