Should CRE firms use an AI agent for commercial real estate?
Use ChatGPT and similar tools where a human reads every output before it matters — market summaries, clause explanations, comp analysis, outreach drafts. Be far more careful with an "agent" that can touch your rent roll, CRM, or outbound on its own. The dividing line isn't capability. It's who approves the action before it leaves the building.
ChatCRE, the AI-for-commercial-real-estate newsletter, used its May 2025 edition to introduce an AI agent built for CRE and to walk through four ways operators can use ChatGPT's o3 model in daily work. That's a reasonable prompt to slow down and ask the question that actually matters for an operating company: what is this thing allowed to do without a person signing off?
This piece takes ChatCRE's four uses at face value — they're genuinely useful — and separates them from the harder case of an autonomous agent with write access to your systems.
What are the four ways to use ChatGPT's o3 model for CRE?
The ChatCRE May 2025 edition frames four uses for OpenAI's o3 model — o3 being the reasoning-oriented model in the ChatGPT family, tuned for multi-step analysis. Each one produces a draft a person then reads. That's the important property: the tool proposes, a human decides.
The four uses ChatCRE describes are:
- Market summaries — condensing submarket activity, absorption, and rent trends into a readable brief.
- Clause explanation — plain-English breakdowns of lease language (an NNN clause, a co-tenancy provision, an early-termination right).
- Comp analysis — organizing and comparing sale or lease comparables to sanity-check a number.
- Drafting outreach — first-pass emails to brokers, tenants, or prospects.
For the rest of this piece, "the four uses" refers to exactly these. Notice the common thread: none of them commits anything. A summary is read. A clause explanation informs a person. A comp set is checked against reality. An outreach draft waits for someone to hit send.
Where do these four uses go wrong?
They go wrong at the exact point ChatGPT is confident and unverified. The four uses are safe because a person reviews the output — but that safety disappears the moment the output becomes the action. The failure mode is a wrong number stated plainly.
Two of the four carry real risk if unchecked:
- Comp analysis invites fabricated or misremembered comps. A model will produce a clean table with a $/RSF figure — rentable square feet — that no property actually traded at. It reads authoritative. It isn't a source.
- Clause explanation can soften or invert a lease term. "Reads right" is not "is right." A misread co-tenancy or termination right is a diligence problem, not a phrasing problem.
Market summaries and drafting outreach are lower-stakes because the whole point is that a human edits them. Keep it that way. The value of these tools is the draft; the judgment stays with you.
Is a CRE "AI agent" the same as using ChatGPT?
No — and the difference is the whole point. Using ChatGPT means you read the output and act on it yourself. An agent is software given permission to take actions — update a record, send an email, move a deal stage — sometimes without a person in the loop. Same underlying model, entirely different risk.
The moment a tool can write to your rent roll, edit your CRM, or send outbound on its own, the review step that made the four uses safe is gone. The question shifts from "is this draft good?" to "do I trust this to act unsupervised on my deals?"
Vantrow's position is direct: propose, never commit. Software should stage the action — draft the broker update, queue the outreach, flag the expiring lease — and a human approves it. Every action lands on an audit trail. That's the difference between a governed assistant and an autonomous agent, and it's the difference that matters when the artifact is your rent roll.
How much of this is CRE actually adopting?
Adoption is wide but shallow: most firms have tried these tools, few have operationalized them. That gap is exactly where governance decisions get made — or skipped.
- According to Altus Group's 2024 CRE industry survey, a large majority of commercial real estate firms report they have piloted or begun using generative AI tools, yet far fewer report production use across teams.
- ChatCRE, per its own newsletter, positions itself as the AI-for-commercial-real-estate publication and used its May 2025 edition specifically to introduce a CRE-focused agent alongside the four ChatGPT uses — a signal of how fast "agent" framing is entering operator vocabulary.
The takeaway: the pilot is easy because ChatGPT drafts and you review. Production is hard because production means the tool acts. Cross that line without deciding who approves, and you've automated your worst day.
What should CRE operators ask before giving an agent access?
Ask what it can do without you, not what it can do. The demo shows capability; the contract and the settings show authority. Before an agent touches your systems, get clear answers to a short list.
- What actions can it take on its own? Read-only is one risk profile. Write access to the rent roll or CRM is another entirely.
- Who approves before something leaves the building? An outbound email to a broker, a change to a deal stage — name the human in the loop.
- Where's the audit trail? Every proposed and approved action should be logged, attributable, and reversible.
- What happens as you add integrations? Each new connection widens what the agent can reach. More integrations mean more surface area to govern, not less to think about.
If a tool can't answer these, it isn't ready for your deal flow. A governed setup answers all four before day one.
Governed assistant vs. autonomous agent — which fits a CRE firm?
For most operating companies, a governed assistant fits and an autonomous agent doesn't. The work — rent rolls, leasing pipelines, lease expirations, broker updates — is high-consequence and low-tolerance for silent error. You want speed on the draft and a person on the decision.
- Governed assistant — drafts summaries, abstracts leases, stages outreach and CRM updates; a human approves; everything is logged. Fits CRE.
- Autonomous agent — takes actions end-to-end with minimal review. Fits low-stakes, high-volume, easily-reversible tasks. Your rent roll is none of those.
The honest answer: use ChatGPT for the four uses today, and hold any "agent" to the propose-never-commit standard before it gets keys to your system of record.
FAQ
Is ChatCRE's AI agent safe to use for commercial real estate?
It depends entirely on what it's allowed to do without you. If it drafts and you approve, it behaves like the four ChatGPT uses ChatCRE describes — reasonably safe. If it acts on your rent roll or sends outreach on its own, ask for the audit trail and the human approval step first.
Can I trust ChatGPT's o3 model with lease clauses and comps?
Trust it to draft, not to be right. Clause explanations and comp analysis both produce confident output that can be wrong — a misread termination right or a comp no property actually traded at. Read every number and every clause against a real source before it informs a decision.
What's the difference between using ChatGPT and running an AI agent?
Using ChatGPT means you read the output and act yourself. An agent is given permission to take actions — updating records, sending emails — sometimes unsupervised. Same model, different authority. The review step that keeps ChatGPT safe disappears when an agent acts on its own.
What does "propose, never commit" mean for an AI tool?
The software stages an action — a drafted broker update, a queued outreach email, a flagged lease expiration — and a human approves it before anything happens. Every action is logged on an audit trail. It keeps speed on the drafting while keeping judgment with a person.
Which CRE tasks are safe to automate first?
Start with tasks where a human reviews the output: market summaries, lease clause explanations, comp organization, and first-draft outreach. Keep write access to your rent roll, CRM, and outbound behind an approval step until you trust the tool and can see everything it does.