Habit 1. Let it ask first
Treating it like a search bar gets you a generic answer. Let it ask you questions first.
Before you start, what do you need to know to do this well?
Becoming an AI-Native Operator
Build your first AI agent
Every prompt from the talk, ready to copy. Plus a step-by-step for your first AI agent, the Monday Briefing.
From Ryan Chapman, Co-Founder & CEO, Lumio
I use Claude in these examples. ChatGPT has its own versions of most of what you'll see.
Jump to the Monday BriefingStart here
Three things turn a chatbot into an AI agent
- Context. So it knows your business.
- Access. So it can act in your systems.
- Schedule. So it gets to work on its own.
Do them in this order. Each section below has the prompts for that step.
1 · Context
A new chat starts with an empty desk and your question. Anything it didn't learn in training has to be on the desk. You'll fill it in once, then every chat starts out knowing your business.
Step 1. Teach it who you are
Don't write it yourself. Let it interview you. Answer out loud if you can (a dictation app like Wispr Flow makes this much faster).
Interview me one question at a time about my role, my company and my goals. Then write it up.
Step 2. Save it as a skill
A skill is know-how Claude uses automatically, in any chat, whenever it's relevant. You do this once.
Turn this into a skill that you use whenever you're working on anything for my business.
Tip: keep the master copy in a Google Doc. When things change, update the doc and ask Claude to update the skill.
Step 3. Teach it how you run things
You probably already have a playbook: SOPs, a manager manual, your late fee policy, your lien timeline. Upload it as-is.
Turn these into a skill called Operating Playbook that you use whenever I ask how we run our facilities.
No written playbook yet? Use the interview trick.
Interview me one question at a time about how we run our facilities: move-ins, late payments, overlocks, liens, move-outs and reviews. Then write it up as our operating playbook.
Step 4. Anything you explain twice, make it a skill
Look at my past chats. What do I keep explaining that should be a skill?
Step 5. Let it learn as you work
Memory picks up the exceptions that never make it into a doc. For anything important, just tell it.
Remember this: [the exception, e.g., the RV spots at Elm St get overlocked on day 15, not day 11].
Every few weeks, check what it knows and correct anything that's wrong.
What have you learned about me and my business? List it so I can correct anything that's wrong.
2 · Access
To do work, it needs your systems. Connectors (built on MCP, the plug between AI and your software) let it read from your systems and act in them. Each connection is a short list of things it can read and things it can write.
Step 1. Start with two: Gmail and Google Calendar
In Claude, go to Settings → Connectors and turn on Gmail and Google Calendar. That's where most of your day lives. Add Google Drive next.
Step 2. Point your playbook at the live doc
Once Drive is connected, your team can edit one copy of the playbook and the AI always reads the latest version.
Update my Operating Playbook skill to read the latest version of [Google Doc name] in my Drive before answering.
Step 3. No plug for your FMS yet? Use email.
Most management software can email a daily or weekly report. Schedule that report to go to your Gmail, and your AI can read it. That works this week, with no integration.
- In your management software, schedule your daily or weekly report to email to your Gmail address.
- In Gmail, create a filter that labels those emails (e.g., "FMS Reports").
- Tell Claude where to find them.
My management software emails a report to my Gmail every [day/week] with the label [FMS Reports]. Read the latest one and tell me what stands out.
Step 4. Ask every vendor one question
"Does it plug into ChatGPT or Claude?" More are adding it.
Watch it work.
Try this once Gmail and Calendar are connected:
Prep me for today's meetings.
3 · Schedule: your first AI agent
Agents wait to be asked. A schedule lets it get to work on its own. Give it a job, the access it needs and a time to run.
The Monday Briefing.
Every Monday at 7 AM, it reads your FMS report and Google review alerts, tells you which autopays failed, which reviews need replies and why occupancy changed, and drafts the reminders and replies for you.
- Context
- your Operating Playbook skill
- Access
- your FMS report and review alerts in Gmail
- Schedule
- every Monday, 7 AM
Set it up in six steps
- Do the Context steps above. At minimum, your Operating Playbook skill, so the drafts sound like you and follow your rules.
- Connect Gmail. Settings → Connectors.
- Get your FMS report into Gmail. See "No plug for your FMS yet?" above.
- Get your Google review alerts into Gmail. Turn on email notifications for new reviews in your Google Business Profile.
- Paste the prompt below. Claude sets itself up as a scheduled task and may ask you a few questions first. Answer them.
- Review the first briefing. Tell it what to change. It gets better every week.
Every Monday at 7 AM, read my management software report and Google review alerts in Gmail. Tell me which autopays failed, which reviews need replies, and why occupancy changed at any facility. Draft the reminders and review replies, but don't send anything without me.
It drafts. You decide. Nothing goes to a tenant until you send it.
One more: the Meeting Manager
Set yourself up to prep me every morning and draft follow-ups every evening.
Tip: connect a meeting recorder like Granola and the follow-ups get specific. Always get consent to record.
4 · Working with it: three habits
Manage it like a sharp new hire. You wouldn't toss them a one-liner and sign whatever they hand back.
Habit 2. Think first, then hand it off
Jot your own take first, even three bullets. Let it build on your thinking. You make the final call.
Here's my take: [your 3 bullets]. Build on it, tell me where you disagree, and what I'm missing.
Example: before you ask for a rate-increase plan, write down who you think will leave and why. Then let it test that.
Habit 3. Make it push back
It's trained to agree with you. Tell it who to argue as.
Challenge this as [a skeptical owner / a tenant / my toughest competitor / my attorney]. Give me the strongest argument against it.
Make it permanent:
Add this to my skill: always end with the strongest argument against what I'm doing.
Then check what it says against something real: your numbers, your FMS, a person.
Which habit do you skip most?
Look at my past chats. Which of these three habits do I skip most: letting you ask first, thinking first before I hand it off, or asking you to push back?
5 · Aim it higher
Once it works, the limit is how ambitious your asks are. This prompt has the AI rewrite your request into what an expert would ask for.
I want a deeper version of this task, not a bigger one. Before you start, answer these: 1. What's the real goal? 2. What would a top pro do that I'm not asking for? 3. What raw data would they start from? 4. What methods and tools would they use? 5. What, besides you or me, would prove it's right? Then rewrite my prompt to ask for that.
Example (before → after)
- Before
- "Write our rate-increase letter."
- After
- Pull 12 months of ledgers and move-outs. Find who leaves after an increase. Set increases tenant by tenant. Check against the next 60 days of move-outs.
- The real goal
- more revenue without more move-outs.
"Deeper, not bigger" prompt adapted from Felix Rieseberg at Anthropic.
6 · In front of your tenants
The same three ingredients work on your phones. Context is your playbook plus the tenant's account. Access is your FMS: payments, gate codes, move-ins and move-outs. The schedule is the phone ringing, day or night.
Six questions to ask any AI vendor
- Is it built from my playbook, or a template everyone gets?
- Does my team control the playbook, and can we change it ourselves?
- Does it learn from the calls it transfers? How fast does it improve?
- Can it handle the hard calls: multi-unit payments, overlocks, lien timelines, move-outs?
- Does it finish the job in my FMS, or just take a message?
- Does it follow the same rules on phone, text and web?
Then ask them to show you, on a real call.
Lumio
Your playbook, answering your phones
Lumio turns your playbook into AI agents that answer calls, texts and web chats for your facilities, and finish the job in your FMS. When it doesn't know, it hands off to your team instead of guessing.