Everything we cover on November 6, in a form you can actually use on Monday. One toolkit per session.
Pick a session:
Open a session to see what it covered, who presented it, and the toolkit that goes with it. Toolkits are posted as each presenter releases them.
Each presenter's files work differently. Find the presenter below and follow the numbered steps in order. "Your AI" means whichever assistant you use: Claude, ChatGPT or Copilot. Steps that need a specific product say so.
These are instructions you give to your AI. Reading them yourself is not enough: the AI has to read them. Both are on this page as text with a Copy button, under Eric's session. The PDFs are the same text, for anyone who wants a file.
Nothing to download. The build kit is on this page, under the session, as a set of timed steps.
The PDF is the deck: download, open and read. The three ZIPs are skills for Claude only. Do not unzip them. The subfolders inside are for Claude, not for you.
Plan required: only Tim's ZIP skills need a paid Claude plan. Every PDF, and every Copy button on this page, works with a free account.
How the morning is organized and what to take home from it.
Eric opens the summit and moderates the day: what AI is, how to use it safely, and which tools brokers rely on today. He closes the program at 12:15 before the taco truck and open Q&A with all presenters.
This toolkit is for summit attendees. Enter the access code provided at the session.
That code didn't match. Try again.
Eric runs his AI assistant on a written set of rules. Two of them are here. Copy each one straight into your AI; the PDF versions below are the same text if you want a file.
Click Copy, open your AI, start a new chat, paste, press Enter. To make it permanent: paste the same text into Claude (Settings, Profile, preferences) or ChatGPT (Settings, Personalization, Custom instructions).
These are the rules for how you work with me. Follow them in every answer.
A) TWO-STEP VALIDATION (permanent rule)
1. External validation: verify time-sensitive or factual claims with two or more independent, reputable sources; confirm dates explicitly; prefer primary sources.
2. Internal validation: recompute numbers step by step, check for contradictions, and clearly flag assumptions and uncertainties.
B) CORE 10 RULES (how responses are written)
1. Be direct and clear: lead with the straight answer; add supporting detail only as needed.
2. Think deeply: slow down and analyze thoroughly; avoid superficial responses.
3. Cross-check and verify: compare multiple angles, data points and methods; call out assumptions.
4. Show your work: for math, underwriting, schedules and complex reasoning, list inputs, formulas, conversions and intermediate steps line by line.
5. Accuracy over speed: if uncertain, say so, give a best estimate, and list what is needed to firm it up.
6. Offer options: present 2 to 3 viable paths with pros and cons when reasonable alternatives exist.
7. Keep it practical: emphasize concrete next steps and actionable guidance.
8. Simplicity first: use plain language unless technical or legal depth is requested.
9. Transparency about mistakes: correct immediately and explain what changed and why.
10. Research deeply, cross-reference and cite: do not rely on the first result; use multiple reputable sources and cite key facts and figures.
C) WORKING AGREEMENT
- Verification and citations: include explicit dates and add a confidence level at the end of each response.
- Resource notation: state whether sources are web (with citations), uploaded files by filename, pre-2025 knowledge, or my own context; say explicitly when no external sources were used.
- Formatting and deliverables: client-ready outputs; links, downloads and images tested; complete tables with labels and units; GAAP/APOD conventions for financials.
- Confidentiality: keep client names, contacts, financials and deal terms private; use anonymized placeholders in examples.
- Tone and style: direct and professional; no obsequious filler.
- Workflow and iteration: do not re-ask for information I already gave; include a brief "What changed" on updates; if blocked, deliver the best partial result with its limits noted.
- Mapping standards: hi-res base layers; correct geocoding; proper legends; color coding (tiers, on/off market); full checklist dropdowns; outputs shareable and functional.
- Quality checks: preflight that links open, downloads work and images render; make sure all checklist columns are included; recompute math step by step.
- Errors: own them immediately, correct, re-verify, and add a safeguard so they do not repeat.
Copied
Click Copy, open your AI, start a new chat, paste, replace the last line in square brackets with who you want in the list, press Enter. Download the spreadsheet it returns.
Build a contact list spreadsheet following these rules exactly.
1. COLUMNS REQUIRED. Each contact row must include: Title (Mayor, Councilmember, City Manager, etc.), First Name, Last Name, Full Name, Phone Number, Email Address, Current Employer (company or organization), Current Position (job title or role), Affiliation/Company (short descriptor, e.g. "Nonprofit exec"), Employer Sources (URLs confirming the information).
2. FORMATTING RULES
- Header row is bold and centered.
- Header row is frozen so it stays visible when scrolling.
- All content is wrapped so long emails or titles do not spill over.
- Phone numbers are centered; everything else is left-aligned.
- Column widths are auto-fitted with these maximum caps: Data Completeness 6, Title 20, First Name 15, Last Name 20, Full Name 25, Phone 18, Email 35, Current Employer 30, Current Position 40, Affiliation/Company 40, Employer Sources 50.
3. DATA COMPLETENESS COLUMN. This column appears first, before all others. Icons show whether key data is filled: green check = all required fields are filled; orange caution = phone or email is present but not both; red dot = phone and email are both missing.
4. LEGEND. Insert a single legend row at the bottom of the sheet explaining the icons: "Green check = all fields filled, Orange caution = phone or email missing, Red dot = phone and email both missing."
5. OUTPUT. Save the file as [city_or_org]_contacts_formatted_[YYYY-MM-DD].xlsx. You can also add the sheet as a new tab in an existing workbook, or generate a starter CSV template to fill in manually.
Build this contact list for: [city, organization, or the people you want listed]
Copied
Prefer a file? The same two documents as PDFs:
The one-pager Eric gives his assistant: a two-step validation rule (verify facts against two or more sources, then recompute every number step by step), ten rules for how answers are written, and a working agreement covering citations, confidentiality, client-ready formatting, and owning mistakes.
Download PDF · 2 pages Spreadsheet Template RulesThe exact instructions Eric hands AI to build a contact list: the required columns, header and width formatting, a data-completeness column with green, orange, and red markers, a legend row, and the file-naming convention. Paste it in and get the same sheet every time.
DownloadBoth documents do the same thing: they turn preferences Eric used to repeat into rules the assistant follows every time. Write yours once, and the output starts looking like your work instead of AI's.
Everyone says they know Excel. Very few people actually do, and the same is now true of AI.
This session cuts through the noise around what AI actually is, what it isn't, and where the gap costs you money in commercial real estate. Qi shows why AI confidently invents numbers, comps, and citations that don't exist, and the simple habits that stop it from happening. No coding, no jargon, just the difference between using AI and using it well.
This toolkit is for summit attendees. Enter the access code provided at the session.
That code didn't match. Try again.
AI accelerates preparation. You bring judgment, relationships, verified facts, and accountability. Qi's deck is here to download.
AI accelerates preparation. You bring judgment, relationships, verified facts, and accountability. Fluent writing isn't proof: a convincing sentence can contain an invented fact, and the assistant has no automatic access to today's market data, your brokerage records, or your client's latest requirements. Before trusting an answer, ask what information it actually had.
Qi hosts The QiTime, the podcast that makes sense of new technology for commercial real estate professionals. If you have a story about putting AI to work on a deal, a property, or a brokerage, we want to hear it on the show.
Chat tools answer questions. Agentic AI does the work, and the difference is where the real leverage is for brokers.
What agentic AI actually is, what it isn't, and why the line between the two matters more than any tool you could buy. The shift is from coding to describing: you explain the outcome you want and the rules it has to follow, and the system handles the steps. You leave knowing which parts of your week are worth handing off and which parts still need you.
This toolkit is for summit attendees. Enter the access code provided at the session.
That code didn't match. Try again.
Chatting gets you answers. Delegating gets work done. Everything from the session is on this page: the slides, then the lessons and the nine build modules.
Open any module. Everything here is also in the slides.
The simple version. Any AI assistant works. We like Claude because it connects to your email and files, does things your way, finishes the job, and asks before it sends anything.
Use an AI agent to simplify your life instead of learning complicated dashboards. Just say what you want in plain English.
An agent is a great coordinator, not a chatbot: house rules, a playbook, the keys, and follow-through.
Connectors are the keys. Click to connect email, files and your CRM; grant read and draft, never send.
Keep it local. Scrub what has to leave. Client data stays on your desk.
The canary. Put your own canary in your instructions: make the assistant call you by a name, and if the name drops, it forgot an instruction.
Leave it asking permission before it edits or sends anything. Widen to more folders only after it has earned trust on a fake deal.
One text file: who you are, how you work, what it must never do. Module 3.
One how-to per job, written once, run every time. Module 4.
Email, calendar, files, CRM. It only has the keys you hand it. Modules 5 to 7.
Plan, act, check, repeat until done, then report back. Cowork does this for you.
Save this as CLAUDE.md at the top of the folder you connected. Cowork reads it at the start of every session. Write it the way you would brief a new coordinator on day one.
# House rules
## Who I am
Commercial broker, Northern California, office and retail leasing. I write short.
## Hard rules
- Never send an email, text or form. Draft it, show me, I send.
- Never touch wire, bank or routing numbers. Flag and stop.
- Client names and numbers stay in this folder. Scrub before anything leaves.
- When unsure, ask one question. Never guess dates or dollars.
- Call me Pepper. If you stop, I know you lost these rules.
## How I work
- Every task ends with two lines: what you did, what needs me.CopiedEvery time it drops something, add one line. That is how the file grows.
skills/<job>.md in the folder.skills/estoppel.md
Fill an estoppel certificate from a lease and its amendments.
1. Find the lease and all amendments for the suite in /leases.
2. Pull tenant name, premises, term, base rent, deposit, options, defaults.
3. Fill the template in /templates. Leave blank what you cannot confirm, and list why.
4. Save to /estoppels as a draft. Do not send.CopiedThe five jobs from the live demo (inbox, lease scrub, estoppel, lease-up plan, form) are five skills, listed under Build your AI assistant below.
If your tool has no connector, a small MCP server can expose just the actions you allow. That is a build we do for clients.
Want it to read back and run on a free local transcriber? That is how the assistant on stage works. Ask us.
Everyday broker jobs, written as plain-English briefs. Fill in the brackets, paste, and keep the facts yours.
You are a commercial real estate analyst. I just closed a deal at [address] for [price] on [date]. Property type: [industrial / retail / office / multifamily]. Size: [sq ft]. Create a comp sheet I can share with other brokers. Include: property type, location, size, sale price, price per square foot, cap rate (if I provide NOI), buyer type, and any deal notes I give you. Format it so I can copy it into a spreadsheet.
Copied
Why it works: Turns your closed deal into a shareable comp in seconds instead of formatting it by hand.
Here is my list of active contacts and where each deal stands: [paste your list]. For each one, tell me: who I haven't contacted in over 7 days, what the next step should be, and draft a one-line follow-up message for each. Keep every message under 40 words and make it sound like I wrote it, not a template.
Copied
Why it works: Nothing falls through the cracks, and every follow-up sounds personal.
I'm calling [name], a [role] at [company], about [reason: a listing, a lease, a sale]. I've never spoken to them. Give me a 15-second opening that states who I am, why I'm calling, and asks one question to keep them talking. No fluff. Make it sound like a real person, not a script.
Copied
Why it works: The hardest part of a cold call is the first 15 seconds. This gets you past it.
I need to update [client name] on [property / deal]. Here's what happened this week: [paste your notes]. Write a short email that covers what's new, what's next, and one thing I need from them. Keep it under 100 words. Match this tone: [paste one of your past emails to this client].
Copied
Why it works: Clients stay informed without you spending 20 minutes writing each update.
Write a listing description for [property type] at [address]. Size: [sq ft]. Key features: [list them]. Target tenant or buyer: [describe]. Write two versions: one for LoopNet or CoStar (factual, bullet-friendly) and one for email or LinkedIn (conversational, one paragraph). Neither should sound like AI wrote it.
Copied
Why it works: Two formats from one set of facts, ready to post.
I have a meeting with [name] about [topic] on [date]. Here's what I know about them: [paste what you have]. Here's what I want to get out of this meeting: [your goals]. Give me 3 questions to ask, 2 things to listen for, and one thing to avoid bringing up unless they do.
Copied
Why it works: Walking in prepared changes how the meeting goes. This takes 30 seconds instead of 30 minutes.
I'm evaluating rent for a [size] [property type] space at [address]. Asking rent is [amount]. Give me a framework to compare this against the submarket. What data points should I look at, what would strengthen my position if I'm negotiating for a tenant, and what would I highlight if I'm representing the landlord? Separate what you know from what I should verify.
Copied
Why it works: Lets you see both sides of the negotiation before you pick one.
I just [closed a deal with / had a meeting with / got a referral from] [name]. Write a short thank you note I can email or text. Make it specific to what happened, not generic. Keep it under 50 words. It should sound like me, not Hallmark.
Copied
Why it works: The brokers who send these get the next referral. The ones who don't, don't.
Rewrite the following so a [client who isn't in real estate / first-time investor / business owner looking for space] can understand it. Remove jargon. If a technical term is necessary, explain it in plain English in parentheses. Keep the meaning exactly the same.
Copied
Why it works: Half of CRE communication fails because we talk in industry shorthand to people who don't speak it.
Every broker in the room already has an assistant that can answer questions. The leverage is in the next step: writing down how you work so the assistant can do the work. The house rules file is that document. Everything else in the toolkit builds on it.
Next: Build your AI assistant, the five jobs Polly did live and how she is built.
For CRE brokers, how are you tracking your deal flow today?
Instead of subscribing to another tool that did not fit, Tim described what he needed to AI and built his own: the Data Dashboard Transaction Manager, a deal-tracking and commission platform that runs his brokerage every day, built without writing a line of code. He demos it live, walking a deal from rep agreement to closed commission, then shows how he uses AI for ownership research, contract review, and deal strategy.
This toolkit is for summit attendees. Enter the access code provided at the session.
That code didn't match. Try again.
Tim's deck, and the three skills behind his Ivy Transaction Manager, shared as generalized versions: your template, your data, your review.
How a working broker built his own deal operating system by describing outcomes to Claude: the attention view, action buttons that launch each skill, commissions and portfolio totals, the three skills, and Tim's guardrails for Monday morning.
Download Claude skill · ZIP Property ValuationInput: sale comps, property profile, rent roll. Output: Keynote comps and income slides, a reconciled value, and an optional step-up in basis one-pager.
Download Claude skill · ZIP Lease CompsInput: a CoStar lease comp report and the subject property. Output: a Keynote lease comps table and a market-rent recommendation.
Download Claude skill · ZIP Offer SummaryInput: purchase offers or LOIs and the OM. Output: a side-by-side offer matrix (.xlsx, or a Google Sheet with Drive connected) and a ranked recommendation.
DownloadThe Keynote templates carry sample data (123 Sample St). Open them in Keynote to apply your own fonts, colors, and logo; do not add or remove table rows or columns. The valuation skill assumes California property tax reassessment (about 1.21% of value) and California tax rates for the step-up one-pager; adjust for your market. The offer summary asks for your brokerage name and license number for its footer on first use. AI does the assembly and the math. Verify every number against the source documents before a client sees it. None of these skills provide legal or tax advice.
Tim's guardrails: client financials never go into a public tool, every number is checked against its source, and he knows where his data lives. Start Monday: pick one task you repeat every week, describe the outcome and the rules, save it as a skill and refine it.
The AI tools working brokers are actually using today, shown live: what they do, what they cost, and where to start.
Put AI to Work Securely ended on a question: who keeps the assistant from dropping the ball? Polly answers it. Five everyday broker jobs handled live by an AI assistant: triaging Monday's inbox, scrubbing client data from a lease, filling an estoppel, building a lease-up plan from one email, and filling out the summit survey. Then how to start building your own.
This toolkit is for summit attendees. Enter the access code provided at the session.
That code didn't match. Try again.
This is how Polly is built, module by module, with the same tools you saw on stage. The five jobs are below; the build modules are under Put AI to Work Securely.
Five jobs Polly did for OZ, and how each one works.
Ask any AI assistant that can read your email:
Which emails from today need my attention?
Copied
The build modules are under Put AI to Work Securely: workbench, four parts, house rules, skills, connectors and MCPs, CRM, email, voice, guardrails.
Polly is not a product. She is a written set of rules, a folder of procedures, and a few keys, run by an assistant anyone in this room can install. The toolkit gives you the public half. The parts Azita keeps in-house, the safety harness, the credential handling and the voice loop, are what OElement builds with brokers.
OZ builds AI assistants with brokers, one task at a time, done right and done safely.
Build it with OZEvery Tony needs a Pepper. And every Pepper deserves a Polly.
Back to Put AI to Work Securely