17 questions came up live. Paraphrased from the actual transcript. Want more depth on any of these? Ask the AI on this page — it has the full context.
Have you gotten better at prompting over time?
Mixed. The models are smart enough now that less prompt engineering matters. The bigger lever is volume and frequency — Connor uses voice transcription so he interacts with AI dozens of times a day, which is where the skill compounds.
How much of your AI use is your company vs personal experimentation?
Most hands in the room said "only sanctioned tools." Few said "experimenting on my own." That's the gap. Personal experimentation is how you find the workflows worth bringing back to IT.
How do you store the lessons learned so AI can read them?
Two layers. Human-readable: the actual reports and transcripts, indexed in a document library with links back to the people and projects. Machine-readable: a vector database where each chunk of text is encoded so the AI can semantically search it, with pointers back to the source.
How would I get coworkers to actually upload their work?
Show them the prototype working on your own data first. Once they see "I asked one question and it pulled from 50 PDFs in two seconds," the upload friction becomes worth it. Sell the magic, not the system.
Can AI log into different platforms and pull data on my behalf?
Yes — brand new. Two paths: API connectors (every modern software has one) and browser-controlling agents (Comet, Atlas). If a human can see it, AI can see it. The 26-property research example: yes, end-to-end automation. Specify the workflow once, run it forever.
What's a "skill"?
A pre-packaged workflow inside the chat. Multi-step recipes the AI can execute end-to-end without you re-prompting each step. Claude calls them Skills. Other tools call them Agents, Custom GPTs, Gems. Same idea.
How do I separate the right answer from the wrong answer?
Two heuristics. Don't pipe AI directly into business systems — always have a review step for anything high-stakes. And match AI quality to reviewer expertise: a senior teaching AI their job produces high-quality output; a junior with no domain knowledge produces garbage. Teach AI the way you'd teach a new hire.
What level of subscription do I need to protect my company's information?
Three tiers: self-serve consumer (data may be used for training — don't paste anything sensitive), business plan (better protections), enterprise (explicit "we won't train on your data" terms, plus HIPAA/SOC 2 as needed). Enterprise is what your IT department negotiates.
What's the right tool inside the Microsoft ecosystem?
Lean in, don't go around it. Outlook + Copilot for email and document work. Power Automate for connecting things — most people don't know they already have it. Email yourself things if Copilot can't reach them directly. Don't try to wire ChatGPT into Outlook. Use Copilot well first.
Is the AI workforce reduction timeline (18–36 months) real?
Probably yes for data aggregator and research roles. Software replacement is more visible than headcount replacement. Most companies aren't reducing — they're not hiring as fast. The shift: think less in terms of roles, more in terms of workflows. Offshore teams doing repetitive knowledge work are most at risk.
How do you handle younger employees using AI poorly?
Two leverage points: keep the standard high (they may not know what "right" looks like — show them the bar), and remember domain expertise is the multiplier. A senior with AI > a junior with AI by a wide margin, because the senior knows when the output is wrong.
How do you respond to AI vs sustainability pushback?
Connor's frame: if AI makes you more effective, less time wasted is generally good. From the room — ULI speakers said "the people getting left behind aren't the ones who lack proficiency, they're the ones who refuse to try."
Can AI generate LinkedIn posts on autopilot?
Technically yes. The algorithm punishes obvious AI content though, and the posts that perform are usually unpredictable. Connor's best post is about staying at a hotel next to a gym — couldn't have planned that.
Are you using AI-generated headshots?
Yes. Some firms (with attorney pushback overridden by the COO) are moving entirely to AI headshots: upload 10 photos, get business attire and weight adjustments for ~$10. Replaces $1K+ photoshoots. Connor's wife can tell. Nobody else can.
What's the best resource to learn AI?
YouTube (things change monthly). Coursiv (coursiv.io) covers ChatGPT, Claude, Midjourney, Jasper, Deep Seek. Use the AI to teach itself — ask it "what can you do that I don't know about?" Try Claude Cowork if you have access — it's the in-between layer between chat and code.
Are people actually paying for tokens?
Yes — tokens are becoming a real line item. One attendee just renewed for her team and is rethinking the IT budget around tokens and credits. Cheaper than vendor alternatives, but tracking them at scale is new territory.
Final unlock?
Record everything. Internal meetings, property walks, idea streams. Worst case: an Apple Voice Memo. Transcripts become a continuous data source AI can mine for content ideas, missed action items, training material, and process improvement.
