Curriculum, not software
A chat answers the question in front of it. An agent carries a defined step of the work: your standing rules, your deal files, a fixed output format, and a person signing off. Here is what that is in CRE terms, and how to build one.
By Zed Truong, Co-founder, Pacific Software VenturesPublished · Updated
An AI agent for commercial real estate is Claude given tools, your deal files, and standing instructions, so it can carry a defined workflow, screening a deal, abstracting a T-12, or drafting an IC summary, rather than answering one prompt at a time. A chatbot answers questions; an agent owns a step of your pipeline and returns output in a format your team reviews. Most products sold as CRE agents are rented software, so the buy-versus-build question is really rent-versus-own. PSV builds CRE AI on Claude by Anthropic and teaches operators to build agents of their own.
Agent vs chatbot
You ask, it answers, and the context ends with the conversation. Useful, but every deal starts from zero.
It carries one step of your pipeline with standing rules, access to the real files, and an output format your team reviews.
Everyone selling AI agents for CRE is selling software. But an agent is not a product category, it is a capability, and one a CRE professional can learn. Where agents sit in the wider toolkit is covered in What is CRE AI.
The stack
Claude by Anthropic, the reasoning underneath. Well-suited to long deal documents and detailed, multi-step instructions.
Your rules, written down once: what to screen for, which assumptions to apply, what the output must look like.
Your deal files. The OM, T-12, and rent roll for the asset in front of you, so the answer is grounded in your own documents and points back at the page it came from.
What the agent may touch: files, spreadsheets, data connections. The scope you grant defines what it can and cannot do.
A person signs off. The agent drafts, cites its sources, and stops. Your team approves what moves forward.
The model is the least of these decisions; the reasoning behind PSV's choice is in Claude vs ChatGPT for CRE. The layers that make or break the agent are yours: the rules, the files, and the gate.
This stack is exactly how the CRE AI Institute teaches members to build their first agent, on a real practice deal, a multifamily asset called Desert Grove.
Rent vs own
Most agent products rent you a black box: your files go into someone else's system and the reasoning stays hidden. An agent you own inverts that.
That is the PSV thesis: cited truth as a system property, not a feature. Operators can build this themselves. Firms that want it built and deployed across the team work with PSV.
Why the Company Brain matters here
The five layers describe the shape of an agent. What separates a capable one from a useful one is whether it still behaves like it works here on the fiftieth deal. A general model can read your rent roll. It cannot know which rollover made your partners walk away last quarter, or that the call still stands. The Company Brain is why an agent you correct today still behaves that way a year in.
A correction outlives the deal it was made on.
The reviewer who overrules an assumption at the gate should not be doing it for the tenth time. Fix it on the deal in front of you and it is still fixed on the one that closes in March.
It behaves like it works here.
The difference between a capable stranger and a colleague is a year of being corrected. That year shows up as habits: the caveats your team always states, the checks nobody is allowed to skip, the number a partner looks for first.
Nothing drifts unless someone decides it should.
One person's clever phrasing on a Tuesday does not quietly become house practice. Behavior moves when a reviewer signs off on the change, and it holds there until someone signs off on the next one.
Two ways from here
The CRE AI Institute, by PSV, walks you through every layer on a real practice deal, from standing instructions to the review gate. 7-day free trial, $0 today.
Frequently asked
Agent teardowns · Weekly
One CRE workflow handed to an agent each week: the rules behind it, the files it worked from, and what the reviewer caught at the gate. Written for operators building their own, not for buyers of software.
No spam. Unsubscribe in one click.
PSV Research Desk
Source-cited reporting on the workflows in this guide.
AI AGENTS
AI agents work best in CRE when they take ownership of a bounded process, make their work visible, and stop at a review gate before a consequential decision or external action.
JUL 12, 2026 / 5 MIN READ
CRE PROPTECH NEWS
EliseAI, which sells AI agents for apartment leasing, resident service, maintenance and renewals, raised $350 million at a $4 billion valuation on September 29, led by Andreessen Horowitz and Bessemer Venture Partners with Ontario Teachers’ Pension Plan participating. It says the money goes to automating more of its customers’ operations, weeks after launching an agent that can act anywhere in its platform.
SEP 30, 2026 / 7 MIN READ
CRE AI EDUCATION
Search interest in an AI course for commercial real estate keeps climbing, and the supply of programs is climbing with it. Most will not change how you work on Monday. The ones that do share a recognizable shape, and it can be evaluated before you pay.
JUL 17, 2026 / 6 MIN READ
Next step
Ten minutes. An honest read on where you stand, the highest-value opportunities, and which engagement fits, if any.