The session everyone showed up for — Michelle Hamilton's full landscape on practical AI adoption in CRE, the prompt framework, and a side-by-side setup for ChatGPT and Claude. ~60 minutes, ~100 attendees.

Michelle's core message

Michelle spent ~30 years in CRE, moved into AI four years ago, built and sold a company (Spark AI Strategy), and now does enterprise AI adoption globally at AnswerRocket. She also sits on one of Anthropic's advisory boards.

AI adoption is a human problem, not a technology problem

If leadership hands AI to a committee, buys licenses, and says "good luck, here are some slides" — adoption fails. People either route around governance with their phones or never learn to actually use the thing.

Start small

Pick one painful, repetitive task that takes more than 15 minutes. That's your first use case. Standardize before you automate.

AI is a prediction machine, not a thinking machine

You're the expert in the loop. It can't guarantee factual accuracy, can't make decisions for you, can't replace your judgment, and won't keep confidential data safe by default.

The four common models

• ChatGPT — she called it "the Kleenex of AI."
• Claude — Michelle's personal pick for ~75% of her work.
• Copilot — the one she gets called in to rescue most often.
• Gemini — solid generalist, especially in the Google ecosystem.

Honorable mention: NotebookLM (Google) — underused, fully contained, great for "talking to" a pile of your own documents.

Data & privacy — the rules she gave

• Upload only what you need. Don't dump a full dataset for a small answer.
• Anonymize. Redact names, dates, and account numbers before uploading.
• Always make the model cite sources and provide hyperlinks — then actually click them.
• Hallucinations are real. If diversity matters to your output, you have to say so explicitly.

The social media rule: if you wouldn't post it publicly on LinkedIn, it doesn't belong in AI. Insider tip: run the first model's output through a second model and ask it to prove the first one wrong.

The prompt engineering framework

The live demo

A three-step lease abstraction in ChatGPT: abstract one lease into a Word table → build an Excel comparison matrix across three leases → generate a one-page executive board brief. Work that normally eats half a day to a full day ran in a couple of minutes. Her companion exercise: time yourself before and after — without tracking hours, you'll never know your ROI.

Setting up the AI

ChatGPT — Projects: your "smart filing cabinet." Name it, upload files that pre-train it (brand standards, voice samples, templates), add custom instructions. Stay in one chat per topic. Canvas mode splits chat and a live editable doc. Choose thinking models over instant almost always.

Claude — Projects: Claude's version of the filing cabinet. Add project-level instructions (tone, role, formatting) and knowledge files; every chat inherits that context. Artifacts is the rough analog to Canvas — documents, tables, and HTML appear in a side panel you can iterate on. The framework works identically: abstract → matrix → brief.

What's next

• Future sessions: agents, deep research, image production, and possibly a dedicated Copilot session.
• A member survey is coming to help shape the group's direction.
• Recaps and transcripts typically take about a week to publish.
• The working group meets right after each session — open to anyone who wants in.

Recurring logistics: 3rd Friday of every month, 10:00 AM ET. Email Ivy at imarguerite@eyeqmonitoring.com · Register · Join the meeting.