I have looked at the same real estate transaction from three very different seats. As an agent, I wanted the next task completed. As a broker, I wanted to know whether it had been completed correctly. As a software engineer, I wanted to know what conditions had to be true before that task was safe to automate.
For most of real estate’s early encounter with AI, those questions could live separately.
Not anymore.
We spent the first wave of generative AI worrying about what the machine might say. Would it hallucinate a fact? Write an embarrassing email? Invent a property feature? Produce something polished enough to make us trust something we should have checked?
There was usually a human sitting between the answer and the action. That buffer is disappearing.
AI is beginning to show up where brokerage work actually happens: inside CRMs, transaction files, document workflows and back-office systems. It can interpret information, prepare work, trigger follow-up and increasingly carry a process forward without someone manually moving every step.
Once software can act, technical capability is only half the question.
The important question is no longer simply, What can AI do? It is: What should AI be allowed to do without us?
Inman Intel reported that 75 percent of respondents in its April 2026 survey primarily used AI through standard chat interfaces, while only 9 percent primarily worked in agentic or integrated environments.
Most real estate professionals are still talking to AI. The technology industry is already teaching AI to act.
Brokerages should define those rules before automation becomes commonplace.
Stop treating AI access like one permission
A brokerage would never give every employee identical access to every account, file, approval and dollar simply because they all work for the same company.
AI should not be treated differently.
Instead of asking whether a tool is simply “approved,” brokerages need to decide what level of authority software receives inside each workflow. I think about that authority in five levels.
Read. AI observes. It can summarize a transaction, retrieve CRM notes, identify missing information or surface a deadline. Nothing changes simply because the AI looked at it.
Recommend. AI develops an opinion. It might identify a missing disclosure, recommend a follow-up or suggest what should happen next. The machine can point. A human still decides whether to walk in that direction.
Draft. AI begins doing the preparation. It can draft the email, prefill information, assemble a checklist, prepare a task or organize a file for review.
Preparation and approval. These should remain two very different things, and that distinction should remain clear in every automated workflow.
Execute. Then comes the meaningful jump. The AI gets a hand on the controls. It can create the task, change the status, schedule the reminder, route the file or trigger the workflow.
This is where a great demo can become a terrible operating system.
Software loves the happy path. Real estate rarely stays on it.
A signature is missing. The closing date changes. Financing changes. Two records disagree. A concession changes the economics of the deal.
Suddenly, the next step cannot be determined simply by asking what happened last time. It requires judgment.
This October, Inman turns its focus to AI and its rapid rise in real estate. During Artificial Intelligence Month, we’re digging into the startups shaking things up and the established players folding AI into their offerings.
The smartest automation may not be the one that knows what to do next. It may be the one that knows when it no longer knows.
Every automated workflow needs an exception path. Stop. Explain the conflict. Preserve the context. Escalate to the right person. Then continue.
Commit. The final level is where an automated action creates a meaningful consumer, contractual, compliance or financial consequence.
Summarizing a transaction is not the same as declaring it compliant. Drafting a client message is not the same as sending it. Calculating a commission is not the same as releasing money. Preparing a document is not the same as deciding the brokerage has satisfied its obligations.
California’s Department of Real Estate made the supervision question tangible in guidance issued in March 2026, warning that using AI to perform licensed activity can raise issues similar to delegating licensed work to an unlicensed assistant.
Different jurisdictions will draw different lines. The operating principle is more universal: Responsibility does not disappear because the click came from software.
Give the brokerage workflow automation a job description
Before adding AI to any meaningful brokerage workflow, I would want five questions answered:
- What can it touch? What records, documents and transaction data does the system actually need?
- What can it change? Reading the deal and changing the deal are different privileges.
- What can it send or finalize? A draft, an internal task and a consumer-facing action should not share the same threshold.
- What sends the file back to a person? Define the conflict, exception or material change that ends the automated path.
- Can we reconstruct the decision? The brokerage should be able to see what the system found, what it changed and where a person stepped in.
That is not bureaucracy around innovation. It is what makes innovation operational.
We tend to describe automation by how many human actions it removes. I think that is the wrong finish line.
The goal should not be to eliminate human clicks. It should be to eliminate the clicks that never required human judgment in the first place. Those are very different ambitions.
The first frees people from repetitive work. The second risks removing people from decisions that still need them.
Real estate spent the first few years of generative AI learning how to prompt the machine. The more consequential era is beginning now.
AI is ready to act. The question is whether brokerages are ready to decide where it stops.
Ryan Persad is founder and CEO at Brokurz. Get connected on LinkedIn and Instagram.
