8:32 AM - Slack inbox
I open my laptop at the kitchen table with black coffee. My Slack workspace pings three times in quick succession. The AI agent has flagged overnight activity: one new demo scheduled for Thursday (good), one customer billing dispute from 2 AM (bad), and a notice that Jenny Park at Park Family HVAC attempted to set up two-factor auth yesterday and it failed. I make a mental note to check on that one myself.
I switch to Gmail. Twenty-two new emails. I archive the thank-you note from Marcus Thompson at Thompson Heating & Cooling - he closed his first two jobs using the quoting template I'd sent him last week - and move on to the harder stuff. There's an email from David Chen at Chen's Climate Control asking about custom integration with his existing invoicing system. That's going to need thought. Not today's problem yet.
I open the Stripe dashboard. Today's revenue sits at $1,247. That's three new signups overnight, probably from yesterday's Google Ads spend. Week-to-date I'm at $8,932. That puts me on track for $4,800 by Sunday if the momentum holds. I do the mental math: roughly thirty-two customers active right now at an average of $150 per month, meaning the churn risk is real. One bad week of follow-ups and it cascades.
9:18 AM - The draft email problem
The AI agent has prepared a cold outreach sequence for five HVAC shop owners it identified from recent Facebook activity. I pull up the drafts in the admin UI. Four of them are solid. The fifth one - directed at a company called Sunrise Mechanical - is using language from a template that feels off. It says "eliminate the headaches of manual dispatch" but Sunrise's Facebook page makes clear they're already using some kind of system. The message would land like we didn't do our homework.
I rewrite it in two minutes. Now it mentions the specific pain they'd raised in a recent post about scheduling callbacks. I approve the other four and leave a note in the UI for the agent: "Sunrise case shows we need to cross-check Facebook URLs against their existing software. Running redundant tools. Let's build that check in."
10:47 AM - The escalation
Susan Rodriguez at Rodriguez Mechanical calls. She's been a customer for six weeks and wants to know why she was charged $150 twice this month. I pull up her account in Stripe. There's the issue: the system processed a payment on the 1st as normal, but then auto-renewed again on the 15th because she'd changed her billing date last month to align with her shop's payroll cycle. The second charge shouldn't have happened.
I reverse it right there in Stripe - add a $150 credit back to her account. Then I do something the AI can't do: I call her back after the refund processes and explain what happened. Not an email. A phone call. She sounds relieved. I tell her I'm adding a note to her account so it won't happen again, and I give her my direct number if it does. That's fifteen minutes of my day that saves a customer who could have churned. She texts me a thank-you after we hang up.
12:08 PM - The metrics hit
Lunch is a sandwich at my desk. I pull up the admin dashboard to check week-to-date pipeline. Current customers: 34. That's up from 31 last Tuesday. But churn is eating us: two cancellations came in yesterday. One was expected - a one-man HVAC guy who said the tool was overkill for his operation. The other was Susan's peer, Linda Vasquez at Vasquez HVAC Services, who said she was "going back to spreadsheets because I know them." That one stings. We'd had no support tickets from her. No complaints. She just decided the friction of learning new software wasn't worth the benefit for a two-person crew.
I flag it in my notes as a segment we need to rethink: shops under three employees might not be our real market. The AI agent is optimizing for demos and closes, but I'm seeing the pattern now. I'll adjust the Google Ads targeting next week.
2:34 PM - The quoting bug and a manual fix
David Chen emails back about the invoicing integration. He's more specific now: he uses QuickBooks Online and wants the HVAC AI quoting tool to sync jobs directly so they show up as invoicing line items. That's a real feature gap. It's not something I can solve in the next week, so I do the honest thing: I email him back, say we don't have that yet, and offer him a 50 percent discount on three months if he wants to stay on board while we build it. I explain in the email that integrating QuickBooks needs real development time and I'm not going to promise it next week. He appreciates the honesty. He says he'll stick around at the discount and we can revisit in Q3.
Then I spend thirty minutes in the admin code looking at a bug the agent flagged: maintenance reminder emails are going out with a wrong technician name in some cases. I find it. It's a field mapping issue in the data sync. Three lines of code. I fix it and push a fix to production.
4:15 PM - The conversation that actually sold someone
My Slack pings. The AI agent has forwarded a conversation with a new prospect named Patricia Ortiz at Ortiz & Sons HVAC. Patricia had requested a demo yesterday but her calendar was chaos. So the agent did something smart: it sent her a five-minute video walkthrough instead of pushing for another call time. She watched it, asked three questions via Slack chat, and the agent answered all three accurately. Patricia just responded: "Okay. I want to start. What's the first step?"
I walk her through a five-minute onboarding, create her account, send her a Loom showing where to input her first job. One new customer. One more email for me to send. I do it manually - no template - because she deserves to know who she's working with for the next year. I tell her I'm here if she hits friction, and I mean it.
5:28 PM - Close-out
I review the day's numbers one more time. Stripe shows $1,911 in new revenue so far (five new signups, one upgrade). Week-to-date is now $10,843. Two cancellations, but three confident new starts, which feels like we're holding ground. My manual email to Patricia got an instant "thanks, ready to go" response.
I look at my to-do list. The David Chen integration thing is sitting there, unsolved. Jenny Park's 2FA issue still needs troubleshooting. Linda Vasquez's cancellation keeps nagging at me - I could reach out, see what it would take to bring her back, but I'm not sure I should chase a two-person operation.
I close Slack and Gmail. The AI agent is doing real work - drafting emails, flagging problems, scheduling demos - but it's an amplifier, not a replacement. I'm still the person who makes the calls, reverses the charges, rewrites the bad emails, and answers the hard questions.
6:01 PM - What's working, what's not
I shut the laptop. Two things are working: the paid ads are consistent, and the phone calls matter. One thing needs fixing: we're not filtering for business size early enough, and we're losing small shops post-sale. One thing surprised me: honesty about product gaps gets more converts than promises. Patricia bought because I didn't oversell. Susan stayed because I actually called her.
Tomorrow I'll adjust the audience filter. Tonight I'm going to email Linda Vasquez and ask what would make this work for a two-person shop. Maybe we build for that. Maybe we don't. But the answer is worth knowing.