Step 1
Name the actions that always wait for you
Before you connect an AI agent to a live deal, write down what it may never do without your yes: send an email or a text to a client, send a signature request, make or answer an offer, delete a record, and anything about money or wire instructions. Give that list to the agent as a rule, and check that the tool can follow it. A tool that writes email as a draft and sends only when you tell it to is safer than one that sends on its own. Before you start, ask your broker which AI tools are approved for client work.
You do
- The list
- Asking your broker which tools are approved
Hand off
- Everything before your yes: the draft, the filled form, the dates
Step 2
Approve the exact message, not the idea
When you tell an agent to "send the buyer an update", it writes words you haven't read yet. Approve the text and the recipient, not the request. A yes to one message isn't a yes to the next one. For a message that goes out on its own, such as a showing confirmation, approve the wording and who gets it once, when you set it up, then read the first few that go out.
You do
- Reading the text and the recipient
- Every yes
Hand off
- The draft, ready for you to read
Step 3
Keep each action on the right deal
A mix-up between two deals is the mistake clients notice first: one buyer's numbers in another buyer's email, or an addendum filed on the wrong property. Name the client or the property in each request, such as "the Coral Way buyers" or "Maria's listing", and take extra care when two deals look alike: two buyers with the same lender, or two listings on one street. Before you approve a draft or a filed document, check the client's name and the address on it. A tool that keeps one record for each client and deal, with its own documents and contacts, makes a mix-up easier to catch.
You do
- Naming the client or the property
- Checking the name and the address before you approve
Hand off
- Filing each document, note, and email with the right deal
Step 4
Check every date against the contract
An agent reads a contract fast, and it can still read the wrong line: the closing date from the first page when an addendum moved it, or a deadline counted from the wrong day. When an agent lists the dates in a contract, check each one against the signed contract and every addendum before a client plans around it. Put the dates you checked on your calendar. When an addendum changes a date, check the new one the same way.
You do
- Checking each date against the contract and its addenda
Hand off
- Reading the contract and listing every date it finds
Step 5
Keep money and wire instructions with people
No AI agent should send, change, or confirm wire instructions. The FBI warns that chat tools can quickly write official-sounding emails with directions to wire funds, so a fake email can look like it came from the title company. Tell every buyer that wire instructions don't change by email, and that they call the title company at a number they already have before they send money. The agent can draft that warning; the call stays with your buyer and the title company.
You do
- Warning every buyer
- Keeping wire instructions out of every chat
Hand off
- The written warning, drafted once for every buyer
Step 6
Read the record of what happened
Use a tool that keeps a record of each step: the draft, your approval, what was sent and to whom, and each document it filed. Once a week, open that record for your active deals and look for anything you didn't expect. When a client or your broker asks what happened on a file, the record answers.
You do
- A weekly look at the record
Hand off
- Writing down each step as it happens
Where Reddy fits
Reddy writes email as a draft in your own mailbox, and you send it, or tell Reddy to send that one draft after you turn on "Send emails on my behalf". A routine sends email only in a send step you approved with its recipient and message. A text to a client goes out only after you say yes to its exact words, a signing request goes out only after you see who signs and say send, and a call waits for your OK on its purpose and its questions.
Reddy keeps one record for each client and deal, and each document and note lands on the right deal. When you send it a signed contract, it lists the dates it finds for you to check. Every action it takes is logged. Wire instructions, and the call to check them, stay with you, your buyer, and the title company.
Questions agents ask
What does human in the loop mean for AI?
A person checks and approves the AI's work at the points that matter, before it takes effect. In real estate, that's before a message reaches a client, before a signature request goes out, and before a client plans around a date.
Who is responsible when an AI agent makes a mistake on a deal?
The AI tool doesn't hold a license; you do. Treat its work like an assistant's: what goes out under your name is yours to check.
Should an AI agent ever send email without asking?
Only a message you approved ahead of time, to recipients you named, such as a showing confirmation. News about a deal, numbers, and anything a client signs should wait for your yes each time.
Is it safe to give an AI agent my contracts?
Use a tool your brokerage approved, one that keeps each client's documents on that client's record and doesn't use your transaction data to train general-purpose AI models. Keep Social Security numbers, account numbers, and wire instructions out of every chat.
Read next
- What is an AI agent? How real estate agents use one, from the first job to a daily routine
- Is ChatGPT safe for client information? Seven steps to keep client data safe when you use AI
- AI hallucinations in real estate transactions: where they hit and how to catch them
- Trust and oversight: approvals and data handling in Reddy