Starting this page as a place to share the journey of learning and building.
No hype. No fake confidence.
Just real progress, real questions, and real growth.
If you’re on a similar path, stick around. We’ll figure it out together.
Question for anyone running AI agents:
Do you want them to tell you everything they did, or only what needs you?
I've gone back and forth. Mine are moving toward "only what needs me."
Which way are you leaning?
Small business owners: what's one thing you'd never hand to an AI agent?
For me it's the hard conversations. A late cleaner, a missed spot, an unhappy client.
Agents can draft. Those ones I want to own.
My ads agent drafted a holiday promo with a "regular price" on it.
A second check compared it to the live booking site. It didn't match. Caught before anything ran.
Rule: any number in an ad gets checked against what the customer will actually see.
Steal it.
Retired one of my automations this week.
It worked sometimes. "Sometimes" was the problem. I had to check it anyway.
An unreliable automation can cost more than none. You still do the work, plus the checking.
Before: one new lead could send the same alert email twice, and each copy woke up an agent.
After: duplicates get filtered first. One alert, one owner.
Less noise in = fewer double replies out.
One morning this week, at the same time:
• Sales agent prepped quotes for 3 new leads
• Ops agent moved a reschedule start to finish
• Ads agent set up holiday tests
Me: reading drafts and tapping approve.
What would you run in parallel first?
My agents each work their own lane.
But anything that needs a yes from me goes through one agent, in one chat.
One place for decisions beats five chats pinging me all day.
Running small holiday ad tests this month with my agents.
They draft the creative, set up the test and tell me in plain English what's working.
I pick what runs and set the budget.
If you could test one holiday offer, what would it be?
Sunday question for anyone running AI agents:
How do you split the jobs?
Mine: one "CEO" for decisions, one for ops, one for sales, one for ads.
Splitting by job made them way easier to fix when something breaks.
What's your setup?
Found a quiet failure in my agent setup.
One agent would say "handed it to sales." The message never arrived. No error, nothing.
Fix: whoever passes the work confirms the other side actually got it.
Sent isn't received. Same with people.
Hot take: an agent that says "done" without proof isn't done.
I ask mine to show me the thing. The link, the sent message, the updated booking.
Takes them seconds. Saves me from finding out on Friday.
Underrated part of running AI agents: naming things clearly.
Two people on my team share a first name. Now each one has a distinct name everywhere the agents look.
People can guess who you meant. Agents shouldn't have to.
Before agents, Saturday was catch-up day. Quotes, texts, next week's schedule.
Now most of that is done or drafted by the time I look.
Saturday is a handful of approvals from my phone. Then it's just Saturday.
What's the job in your business that's 80% prep and 20% decision?
For me it was quotes. Look up the home, run the numbers, write the message.
Now an agent does the prep. I make the call.
What's yours?
This week my agents handled:
• Quotes ready when leads came in
• A reschedule start to finish
• Holiday ad tests
• A price check before an ad ran
I did:
• Tapped approve
• Made the judgment calls
Honest note: a few handoffs failed quietly. Fixed now.
Checked where my AI usage actually goes. Surprised me.
Not the scheduled automations. Most of it is me chatting with my agents through the day.
The set-and-forget stuff is cheap. The thinking out loud is where it adds up.
Where does yours go?
Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data.
Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.
Before: an agent chased cleaners for job photos after every clean. Hit or miss.
After: photos go into the booking app as part of the job. The agent just reads them.
Sometimes the fix isn't a smarter agent. It's moving the step to where the work already happens.