8:42 AM - Inbox triage
I open my laptop at the kitchen table. Coffee's still hot. The Slack alert pinged at 7:15 - three new signups overnight from the SEO campaign. The dashboard shows they came through the "email subject lines for SaaS" landing page. One of them already sent an email to support.
I flip to Gmail. It's Derek Xu from a fitness app called Solstice. He's asking if our AI can handle product copy that ties to user sentiment data. Not a typical request. I flag it with a star and move on. He's getting a free trial first anyway.
The metrics sitting in my Slack notifications are good: 312 active free trials in the system, 41 of them in week two (likely to convert). Monday we hit $4,200 in new MRR. Today is tracking normal - I can see in our Stripe dashboard that two customers paid this morning. That's $158 in recurring revenue just from people who were already onboarded.
But I also see a churn notification. Rachel Kim, who signed up last month and upgraded to the $79 plan in week three, downgraded to the $49 plan yesterday. I make a note to check her account. That's information.
10:15 AM - Copy review and approval
The first real work of the day: I log into our admin dashboard and pull the queue of AI-generated copies waiting for human review. There are eight pieces sitting in "pending_review" status. These are drafts the agent created overnight and this morning - emails, product page headlines, Facebook ad copy. Each one has a confidence score. Six of them are above 0.87. Two are below 0.73, which means they likely need rework.
The first one is from Marcus Webb at a newsletter company. He asked for three variations of an email subject line for a "re-engagement campaign." The AI generated nine options, ranked them, and pulled the top three. I read them:
"Your newsletters are about to disappear - save them now."
"The stories you loved are waiting for you."
"We miss you. $20 credit inside."
These are solid. They're specific, they avoid hype, they give the reader a reason. I approve them and send them to Marcus. The system logs the approval, and an automated email goes out to Marcus saying his copy is ready. This is the part that feels good. The AI did the thinking; I just validated it. Marcus will test these tomorrow, probably use one, and it'll work.
The second one is from Carol Reyes at Reyes Family Practice. She's a dental office owner. She asked the AI to write email copy for a "new patient welcome sequence" - five emails, one for each day. The AI drafted two. They're fine. Not great. They're generic. I read the confidence scores: 0.68 and 0.71. I don't approve. Instead, I leave a comment: "These feel like they could be more personal - what if we reference the intake form they filled out? Or ask about their anxiety level around dental work?" I mark them for rework and the agent will regenerate them using my note.
This is the operator's job: knowing when to approve and when to push back.
11:45 AM - A flagged conflict
I'm scrolling through the support inbox when I see a ticket from last night that I missed. Subject: "Your AI made my copy sound like a robot."
It's from Jay Thompson, a solopreneur selling online courses. He asked the AI for copy for his landing page, got back three variations, paid us $79 last week for the monthly plan, and this morning he's frustrated. He says the copy the AI generated is "too formal for my brand" and that he wasted money.
I read what the AI generated. It's technically good copy - it has a clear value prop, it's scannable, it hits the emotional angle Jay requested. But he's right. It's polished. It's professional. And his brand is... I check his site. His brand is scrappy, first-person, casual. "I hated doing spreadsheets so I built a course about it."
The AI didn't know that. The agent works from the brief, not from immersion. I need to handle this one manually. I write to Jay: "I see the issue. Your copy needs personality, not just structure. Let me generate a new set with more of your voice baked in. Give me 24 hours." Then I pull his original brief, rewrite the prompt to the AI with specific guidance ("casual, first-person, irreverent tone - think indie hacker, not corporate"), and regenerate.
This is the hard part. We can't always get it right on the first pass, and some of my job is managing expectations. But I also don't want Jay churning. If I fix this, he keeps the $79. If I don't, I lose him and he tells two people.
12:30 PM - Lunch and metrics
I grab a sandwich and pull up the metrics spreadsheet. Every Tuesday, I do a week-to-date review:
- Free trial signups (Mon-Tue): 11
- Conversions to paid (Mon-Tue): 3
- Churn: 1 (Rachel Kim)
- New MRR added this week: $6,780
- Total active paying customers: 189
- Trial-to-paid conversion rate this month: 18.2%
The numbers are tracking. We're on pace for $14,200 MRR by end of month. That's where the Year 1 ARR model said we'd be by now. Some days feel like I'm just barely keeping up, but the system is working.
I also check the Linear board where I log feature requests and bugs. Seven open tickets. One is a bug I found yesterday: the AI is sometimes doubling line breaks in email copy, which looks broken when customers paste it into Gmail. It's marked "needs_fix." I add it to today's to-do.
2:08 PM - A customer escalation
Derek Xu from this morning - the fitness app guy - sends another email. His free trial has already generated its first output. He's asking if we can train the AI on his customer sentiment data. Specifically, he has feedback from his app that tells him which features his users are most excited about, and he wants the product copy to emphasize those.
This is beyond the product's current scope. Our AI takes a brief and generates copy. It doesn't integrate with his data pipeline. But it's a good ask, and Derek is probably going to be a high-value customer if he signs up. He's thinking about this deeply.
I email him back: "This is a smart approach and not something we support out of the box yet, but here's what you can do: export your top feature feedback from your sentiment data, paste it into the brief field, and the AI will weight it heavily. You'd need to do that export manually each time, but it'll work. Let me know if you want to jump on a call to walk through it." I also offer a 50% discount on his first month if he commits to the annual plan. This is the work of a small operator: custom-fitting solutions for customers who ask the right questions.
He writes back within 20 minutes. Yes, let's do it. I process his upgrade in Stripe and send him the walkthrough. That's $79 multiplied by 12 months. That's $948 in ARR landing in one email.
4:15 PM - Bug fix and one more batch
I tackle the line-break bug. I log into Linear, pull the ticket, and trace through the code. The issue is in how the email formatter escapes newlines when it returns the AI's output. Twenty minutes of digging. It's a one-line fix in the template rendering. I push it, test it with a fresh copy generation, and mark the ticket closed. I also add a note to the bug: "QA: test with line-heavy email formats before shipping." This is part of running a SaaS: you're not just selling, you're also fixing what breaks.
I then circle back to Carol Reyes at the dental office. Her reworked copy came back from the AI. I read it again. Better. It's got a personal touch now, references the intake form, asks about anxiety. I approve it and send it to Carol with a note: "Rewrote this with your tone in mind. Let me know if it resonates with your patients." She replies 15 minutes later: "This is exactly what I needed. Upgrading to the $79 plan today." Another win. Another customer who gets it.
5:45 PM - Pipeline and reflection
Before I close out, I check the dashboard one more time. Today's numbers:
- Signups: 4
- Revenue added: $307 (Derek's annual commitment over one year; Rachel Kim's downgrade offset)
- Active customers: 190
- Free trials in system: 319
Week to date, we're solid. Next week will probably be quieter - Tuesdays tend to bring signup spikes from the SEO traffic that landed on Monday. The rhythm is becoming clear.
I think about Jay Thompson, the course guy. He's going to get his regenerated copy in 24 hours, and if I do my job right, he stays. I think about Derek Xu, who just became an annual customer because I picked up the phone. I think about the two pieces I approved this morning that are now in Marcus Webb's hands, ready to test. And I think about that line-break bug that would've annoyed someone tomorrow.
This is the work: I'm not writing the copy. I'm not building the code from scratch. I'm the person in the middle, making sure the AI's output matches what customers actually need, fixing what breaks, and keeping the system running. Some days it feels like I'm barely staying ahead of it. Most days I'm not. But most days something good happens too.
I close the laptop at 5:48 PM. Tomorrow will be slower. By Friday, things will pile up again.