Carrie Soave
AI agents can interpret information, use connected tools and carry out defined steps in a business process. In real estate, that can include sorting incoming enquiries, preparing a reply, updating a record or routing work to the right person. What an agent can actually do depends on its integrations, permissions, instructions and the checks built around it.
Start with one process. Give it a clear trigger, an approved source of information and a person responsible for exceptions.
How is an AI agent different from a chatbot or an automation?
| System | Typical role | What to check |
|---|---|---|
| Chatbot | Produces an answer in a conversation | Whether it can access current facts and whether a person must act on the answer |
| Rules-based automation | Executes predefined steps when conditions match | Whether the input meets the conditions and what happens when it does not |
| AI agent with connected tools | Interprets input and selects permitted actions | Which actions it can take, what needs approval and how its work is recorded |
The labels overlap in commercial products. Evaluate the actual behaviour and permissions. A tool called an agent may only draft text; another may be able to change records or send messages.
Which real estate tasks are suitable starting points?
Start where the inputs and outcomes can be checked. Examples include classifying an enquiry, drafting a listing update from verified facts, summarising a call for review or routing an internal request. Each task needs a defined source and an observable result.
For client-facing work, separate preparing an action from approving it. An agent can draft a response without having permission to send it. A transaction reminder can be prepared from an approved milestone record without allowing the agent to change a contractual date. Increase permissions only after the workflow has passed realistic tests.
What would a lead-response workflow look like?
This is a design example, not a client result. A new enquiry arrives. The system checks the contact record, the source of the enquiry and whether a response is permitted. It prepares an acknowledgement using approved information and assigns the conversation to the appropriate person.
If the prospect replies, asks for a person or gives an opt-out instruction, the pending sequence must respond to that event. If the contact's details or the appropriate next step are unclear, it pauses for review. The success condition is a correctly handled enquiry, not simply a message marked sent.
What needs human approval?
Keep approval where an action changes a client's commitments, relies on professional judgement or could create a material error. Examples include advice, contract changes, financial claims and public statements about a property. Agree on the exact boundaries with the person responsible for the business process.
An agent should have only the access its task needs. Record what it read, what it changed and whether a person approved the action. A named owner should be able to stop the workflow, correct the source information and review a failure before it repeats.
How do you know the system is ready?
Test the ordinary enquiry and the inconvenient ones: missing information, duplicate records, unavailable appointments, an interrupted conversation and a disconnected integration. Check the result in the receiving system. A successful tool call alone does not prove the right record changed or that a message reached the intended person.
Use a small controlled rollout before extending the workflow. Record its scope, the tests that passed, the open limitations and who watches it. Keep a recovery path that does not depend on the agent explaining its own error.
Questions about AI agents
Does an agent replace my team? A workflow changes specific tasks. Staffing outcomes depend on the actual work, volume and responsibilities in the business; they cannot be inferred from an agent demo.
Does it work with every CRM? No. Check the particular integration, available actions, account permissions and product limits before promising an outcome.
Can it run all day? A scheduled system can operate outside office hours, but communication timing, permissions and escalation still need to be defined and tested.
Who maintains it? Assign an owner before launch. Software changes, stale instructions and incorrect source data can all change how a working process behaves.
Sources and further reading
The business behind the systems.
From Artificially Real, Carrie’s show documenting the build of an AI-first company. Explore the show.
Start with one process.
Guided implementation or a done-for-you build. Bring the work that keeps getting stuck and the tools your business uses.
Discuss your business ↗