CRE AI Agent Skills for CRE: 4 Ways to Standardize Them
A press release from The AI Consulting Network, carried on Business Insider’s markets wire, announced the expansion of an open-source library to 90 AI agent skills aimed squarely at commercial real estate — underwriting, lease abstraction, due diligence. Check the dateline on the wire item itself before treating it as current, and treat any such release as an announcement, not an audit: read the license, read the actual skill files, and test on your own documents before anything touches a live deal.
The underlying shift is still worth understanding. The unit of AI work in a brokerage is moving from “a clever prompt someone pasted in Slack” to “a versioned skill the whole team runs.”
What an agent skill actually is, without the hype
A skill is a folder of instructions — and often a checklist, a template, and sometimes a small script — that an AI assistant loads when it recognizes the task. Anthropic’s Agent Skills documentation describes the format as a SKILL.md file with instructions plus optional supporting files, surfaced to the model on demand rather than jammed into every conversation. Other assistants have analogous constructs under different names (custom GPTs, project instructions, system prompt packs).
What that buys you is consistency, not intelligence. The model is the same model. The skill is the difference between three analysts producing three differently-shaped rent roll summaries and all three producing your firm’s summary, with your field names, in your order, flagging the same exceptions.
Option 1: A shared prompt library your team maintains
The free starting point. A Notion page or shared doc with your best prompts: “Abstract this industrial lease into these 14 fields,” “Summarize this LOI against our standard terms.” Everyone copy-pastes.
Where it wins: zero cost, zero procurement, immediate. If you are searching for free commercial real estate AI automation, this is honestly most of what “free” gets you, and it is not nothing.
Where it breaks: nobody maintains it, three versions fork within a month, and there is no governance trail showing who ran what against which document. Copy-paste also means client documents move through whatever consumer account the broker happens to be logged into — which is how shadow AI problems start.
Option 2: The AI features already inside your CRE software
Most major listing, CRM and market-data platforms in this sector have shipped some form of AI capability into their products. These are skills too — just ones the vendor wrote and you cannot see. What’s actually available differs by product and plan, and changes fast, so check your own vendor’s product documentation and release notes rather than a roundup article.
Where it wins: the data is already there. A marketing description generator inside your listing platform already knows the square footage, the suite mix, and the photos. No integration work, no data movement, and the output lands in the system of record.
Where it breaks: you get the vendor’s opinion of the right output. If your firm’s offering memorandum has a specific narrative structure, or your abstract template carries fields a national vendor never considered (percentage rent breakpoints by category, say), you cannot change it. You also cannot run the feature across data the vendor does not hold.
Option 3: Adopting an open-source or third-party skill pack
This is the newly interesting option. Public skill libraries — the CRE library above, plus general-purpose skill repos — give you a working starting artifact for jobs like lease abstraction and DCF review.
Where it wins: it is the fastest education available. Reading a well-written lease abstraction skill teaches your ops lead more about structuring AI work than a month of blog posts. And a decent skill file is editable — fork it, swap in your field names, delete the sections that do not apply to your asset class.
Where it breaks: provenance and fit. You do not know who wrote it, whether it was tested on leases resembling yours, or how it handles the ambiguous cases that matter (a CAM cap that references a prior-year base, an option notice window measured in business days). Check the license before commercial use, and never assume a borrowed skill’s field definitions match your accounting team’s.
Option 4: Custom skills wired to your systems through MCP
The full version: skills your firm authors, versioned in a repo, paired with a custom MCP server over your CRM and document store so the assistant can pull the rent roll, look up the comp set, and write results back to the deal record instead of returning text you re-key.
Where it wins: output matches your template exactly, the skill can read your historical deals, and every run is logged against a real identity with real permissions. This is also the only option where a skill can act — update a pipeline stage, file the abstract, start the tour scheduling sequence.
Where it breaks: it is a build, and it introduces a failure mode the other three do not have. The moment an agent with write permissions reads an untrusted document — a lease PDF or LOI drafted by the other side — text inside that document can attempt to steer the agent’s next action (prompt injection). Scope connections read-only by default, give each skill the narrowest tool set it needs, and require human confirmation on any write-back to the CRM or document store. Beyond security, someone has to own the repo, handle auth, and update skills when your template changes. If your firm has no technical owner and no appetite to hire one, a well-maintained Option 1 beats an abandoned Option 4 every time.
A borrowed skill tells the AI how the industry does a job. A custom skill tells it how your firm does it — and that difference is the entire reason your ops team still has to redo the output.
Choosing a tool comes down to the job, not the brand
The most common question — which AI tool is best for commercial real estate — has no single answer, and anyone who gives you one is selling something. Frontier assistants (Claude, ChatGPT, Copilot) are strong at document reasoning and drafting and are where skills live; we compared their day-to-day fit for brokers in this breakdown. Purpose-built CRE platforms win where the data already sits inside them. Deterministic workflow tools still beat AI for anything genuinely rule-based — a date calculation does not need a language model.
A practical heuristic, stated as our opinion: pick the tool by where the data lives and how much the output format matters. High format sensitivity plus data in your systems → custom skill. Low format sensitivity plus data in the vendor’s system → use the vendor feature.
Where skills actually fit a broker’s week
The honest answer to how AI can be used in commercial real estate is task-by-task, not wholesale. Skills fit best where the job is repeated, document-heavy, and has a defined output shape: lease and rent roll abstraction, first-pass underwriting checks, call-for-offers bid grids, tour recaps, standardized follow-up sequences, and property marketing copy. They fit worst where judgment, relationship context, or negotiation posture drives the output — pricing strategy, when to push a seller, whether a tenant’s credit story is believable.
Two rules of thumb people keep asking about
The 2% rule — that monthly rent should equal at least 2% of purchase price — is a widely circulated heuristic that originated in small residential rental investing, not an industry standard. Operationally, the point is output design: if a skill computes it, the output should name the rule, show the inputs used, and label the result as a heuristic rather than returning a bare pass/fail that a reader could mistake for an underwriting conclusion.
The 30% rule in AI is not a defined standard from any recognized body, despite how often it is repeated. People use it loosely to mean “expect AI to handle roughly a third of a task and plan for human review on the rest.” Use it as a mental posture if it helps; do not cite it as a benchmark, and do not build a business case on it.
Modeling the payback without inventing numbers
Do the math yourself with your own inputs. The formula for one skill:
(A × B ÷ 60) × C = monthly recovered cost
Then the part most firms forget: recovered hours only pay off if they get reallocated. An analyst freed from abstracting leases who now sits idle saves you nothing. Decide in advance where the time goes — more BOVs, faster tour turnaround, more prospecting touches.
Our full walkthrough of this calculation, including maintenance cost, is in the brokerage automation ROI guide.
A two-week way to test this
-
Clear the confidentiality question first
Before any client document goes near a borrowed skill, check what your engagement letter, listing agreement, or NDA permits, and confirm which account the files flow through — your firm’s governed workspace, not a personal consumer login. This is the same exposure the shadow-AI problem creates, just with better tooling. -
Pick one task with real volume
Choose something you do at least weekly with a defined output — lease abstraction into your template beats “help with marketing.” -
Time the manual version honestly
Five real runs, stopwatch on. This is your only credible baseline. -
Start from a borrowed skill
Pull an open-source CRE skill for that task, read it end to end, and note every place its assumptions differ from yours. -
Fork it to your template
Replace field names, add your exception flags, add the “say you don’t know” instruction for ambiguous clauses. -
Run it against known-good files
Use five deals you already abstracted by hand, ideally ones where confidentiality is not a question. Score field-level accuracy, not vibes. -
Decide on plumbing
Only if the skill works, ask whether connecting it to your CRM and document store via MCP is worth the build — or whether copy-paste is fine at your volume.
Not sure where to start?
Get a free automation audit: we map your deal pipeline, marketing, and back-office workflows and show you what's worth automating — before you spend a dollar.
Get a free automation auditRelated guides
JLL GPT vs Your Own CRE AI Agent: What to Copy
AI JLL GPT explained honestly — what enterprise CRE AI actually does, and how mid-size brokerages compare off-the-shelf tools vs a custom MCP agent.
Read guide →
AI for Commercial Real Estate Marketing: Agent Skills
Use AI agent skills to standardize listing copy, buyer email blasts, and broker follow-up — what skills are, where they fit CRE marketing, and when to build vs buy.
Read guide →
Commercial Real Estate AI Agents: Connect Claude via MCP
How to give an AI agent governed access to your CRE listing, deal and lease data with a custom MCP server — plus when off-the-shelf AI wins instead.
Read guide →