Brokerages Are Announcing AI. The Shift That Matters Is AI Running the Operations

Three brokerage AI stories crossed my desk within a few days of each other. SERHANT. announced an AI-powered brokerage. Compass made a board-level AI hire. Real Brokerage built an event push around AI. The coverage reads like a category moment, and I agree it is one. I just think the moment that really counts comes later, and it will be quieter.

When I read announcements like these, I ask one question: what does the AI actually touch? The language I read leaned toward the layer a client sees: search, marketing, agent productivity, the experience on top. That work has real value. It also leaves the operating model underneath right where it was.

The week I wrote this, COAST had 25 homes scheduled to close. That's 25 sets of deadlines, inspections, lender conditions and signatures, plus a lot of people waiting on the right update at the right hour. That's where I point AI, because that's where a team actually wins or loses a week.

Quick Facts

• The news: SERHANT. announced an AI-powered brokerage, Compass made a board-level AI hire, and Real Brokerage pushed an AI-focused event, all in the same news cycle.
• The cross-vertical pattern: ServiceTitan became the system home-services contractors run on by owning dispatch, scheduling, invoicing and payment.
• The distinction: AI applied to the marketing layer versus AI running the operations underneath it.
• Operator receipt: COAST had 25 homes scheduled to close in a single week.
• Operator: Donna Gilmore, Director of Operations, COAST brokered by eXp Realty
• Team rank: #1 eXp Mega Icon Team in South Carolina
• Documented: May 4, 2026, in a public, dated build series
• Validation: the build series carries public timestamps that predate this wave of announcements

What the ServiceTitan pattern says

If you want to see where this goes, look at the trades. For a long time, the software a plumbing or HVAC company bought sat around the edges: a booking widget, an ad tool, a review tool. ServiceTitan went after the middle. The call comes in and a technician gets dispatched. They arrive with the job history on their phone and build the invoice on site. The payment clears, and the maintenance agreement renews on schedule. That whole sequence runs through one system.

Once a company runs the work order, the marketing tools plug into it. The company's data, its customer history and its margin end up living there too. The edge tools kept their value, and the middle became the place where the business happens.

Own the work order and the rest follows.

Real estate has its own work order. It's the transaction: contract to close, with dozens of dated obligations in between. It's also the lead that needs a response, the commitment a leader made in Tuesday's meeting, and the per-office numbers someone has to reconcile. Whoever runs that layer ends up holding the part that's hard to replace.

Why the marketing layer goes first

I understand why brokerages start at the top. Marketing AI demos well. On a stage, a listing description writes itself in thirty seconds and the audience gets it right away.

Operations AI barely demos at all. When it works, nothing visible happens. The inspection response goes out on time. The lender condition gets caught early instead of the week of closing. The buyer's agent gets the update before they think to ask. You can't put a closing that didn't slip on a keynote slide.

So the announcements cluster where the demo is, and that's a reasonable business choice. It does mean the press cycle is mostly measuring the layer that's easiest to show, and I'd read the coverage with that in mind.

I wrote about the human side of this in why AI rollouts stall inside real estate teams. The short version: when a tool asks the team to change how they work just to get its output, it usually ends up sitting unused.

What the operations layer does on a 25-closing week

At COAST, the operations run through COASTclaw, the agentic AI system I built and run on OpenClaw. During the week with 25 scheduled closings, this is what it looked like from my chair.

The system watches each file's dates continuously. When something runs late, it gets flagged to the person who owns it with the context already attached. That might be a missing document, an unanswered condition or a deadline that's closer than it should be. The system drafts the updates that need to go out, and it tracks leadership commitments from the week's meetings until they're done. Marketing for new listings keeps moving at the same time, on-brand, without anyone stopping to babysit it.

The team doesn't log into an AI to get any of this. They see the results where they already work: an email, a task, a finished asset, a heads-up before a problem turns into a phone call. That's the standard I hold it to. If the team has to do something extra to get the benefit, I've just handed them another task.

The agents and transaction coordinators on the team still do the real work with clients, and I want to be clear about that. What changes is how much of their week goes to chasing status instead of doing the job.

The receipt that taught me the most

The most useful lesson I've had from running this came from a daily health report, not a win.

The system writes me one every morning. Over time it had grown a long tail of red lines nobody acted on. Some were real and many were stale. The problem was that a truly critical failure looked exactly like the noise around it, so it could sit there and read as normal.

I fixed it at the root. I traced every critical line back to where it came from. The real ones got an owner and a tracked item, and the stale ones got retired. The report now carries a small number of lines, and each one means something.

A launch event can't show you that part of operations AI. When a marketing tool fails, you get a bad caption and you see it right away. When an operations layer fails quietly, a closing date drifts while every screen looks fine. Whether you can trust the quiet is the whole job. Earning that trust is mostly unglamorous work on the boring layers, which I covered in more depth in Part 2 of the build series.

What I'd ask any brokerage announcing AI

I'm genuinely glad these announcements are happening. Serious money and attention on AI is good for this industry, and the large brokerages have distribution and capital I'll never have. If I sat down with the people running these launches, these are the questions I'd want to compare notes on:

1. Which dated obligation in the transaction does your AI own end to end?
2. What does it do on a week with 25 closings that it doesn't do on a slow week?
3. When it fails, who finds out, and how fast?
4. Does an agent have to do anything new to get the output?
5. Once the alerts start piling up, how do you tell a real one from the noise?

I ask my own system the same questions every week.

Where this lands

I'll be straight about my stake. I've been building on the operations side of this question, quietly, inside a working luxury team, and documenting it in public with dates attached. What I built and run at COAST is the first documented team-level production multi-agent agentic AI operations platform in luxury residential real estate. What I care about is simpler than the label: 25 closings can land in one week and the operations around them hold.

The ServiceTitan pattern says the lasting position goes to whoever runs the operations. In real estate that position is still open, as I laid out in where vertical AI stands in this industry. The current announcements put AI on the stage. The next round gets decided in the back office.

Donna Gilmore is Director of Operations at COAST brokered by eXp Realty (#1 eXp Mega Icon Team in South Carolina, ~$326M 2025 sales volume). She is the sole creator and operator of COASTclaw, the first documented team-level production multi-agent agentic AI operations platform in luxury residential real estate, built on OpenClaw. Read the build series: Part 1 | Part 2.