8:42 AM - Inbox triage
I open my laptop in the kitchen with a cold coffee from yesterday. The dashboard is already loaded in my browser. Three new signups overnight. $177 added to this month's MRR. That brings me to $1,804 so far. Six weeks ago I was at zero. Still getting used to that number existing.
The Slack alert channel is quiet. No fires. I flip to Gmail and scan the inbox. There's a new customer message from Carol Reyes at Reyes Family Practice, the urgent kind with a subject line in caps: "AGENT SENT WRONG PITCH TO FORBES." My stomach tightens a little. I read the full email. Carol's startup, a dental practice management tool, tried running a campaign last night targeting health reporters. The agent pulled a journalist who covers hospital systems, not primary care, and the personalized pitch referenced competitor analysis for acute care clinics instead of practices. Carol caught it in the Gmail draft folder before it sent, but she's furious that she had to catch it at all. She paid $59 last month and this is her second campaign.
I make a note in Linear to investigate why the journalist database pull isn't filtering by practice type correctly. But first, I need to talk to Carol.
10:17 AM - Customer escalation
I call her instead of emailing. She deserves that.
Carol picks up on the third ring. I can hear she's stressed but relieved someone's picking up the phone on a Tuesday morning. I walk her through what happened: the agent uses a vectorized search against the journalist database, and there's a gap in the metadata around practice versus hospital coverage beats. It's a real gap. I don't blame her for catching it, and I don't blame the system either. It's just incomplete data.
"Can you run the campaign again?" she asks. "For free this time?"
I think about it. She paid. The agent partially failed. "Yeah. I'm going to rerun the beat matching with better filters and I'll send you three draft pitches to review before we send anything. Give you full visibility. Then we do the campaign, no charge."
She softens. We talk for four more minutes. She tells me she's trying to grow their sales team and earned media is one of three acquisition channels. That context is useful. I tell her I'll send the drafts by end of day.
Hang up. I immediately message her in Slack because she's on the free Slack tier from her trial. Tell her the same thing there. Redundant, but she'll see it fast.
11:30 AM - Fixing the beat data
I go into the Linear board. I create a card: "URGENT: journalist database missing practice-type metadata." I tag it and estimate it as a half-day fix. The database lives in a CSV that I'm currently managing manually, syncing from Muck Rack monthly. I should automate this, but right now I'm the person who does it.
I spend forty minutes going into the database and adding a practice_type field to the twenty journalists we have who cover healthcare. It's boring work. I document it. Then I write a quick script that will catch this field in future Muck Rack exports so I don't have to do it by hand again.
I test the agent against Carol's company profile once more, and the pitch is better. Not perfect, but better. The journalist match is actually relevant now.
I message Carol on Slack with three draft pitches. She responds within eight minutes saying she'll review and get back to me by noon.
12:47 PM - Metrics and the churn notification
I open the Stripe dashboard. It's a Tuesday ritual. I'm looking for three things: new churn, stuck payments, and which plans are converting.
I see a churn notification. Mark Chen, who signed up two weeks ago, just canceled. Reason: "Not enough journalists in my vertical." Mark is in deep-sea fishing equipment, which is... not exactly Press AI's core market. The ICP is tech startups. I remember the signup now. He came through an SEO keyword "press release generator" which cast a wider net than I prefer. I could have talked him through the fit, but the free trial is a trial.
I make a note to tighten the onboarding flow. Add a qualification question: "What industry are you in?" If someone answers with a niche that's three steps removed from tech or enterprise software, I'll put a note in the onboarding email saying Press AI works best for software startups. Not a rejection. A heads-up.
Net: I'm still adding subscribers. But churn is real at seven percent this month. That's higher than I want.
The Stripe dashboard also shows me that I have three invoices sitting in draft status. One customer's card got declined. I'll handle those after lunch.
I grab actual lunch. Thirty minutes away from the screen. Read some tweets about SaaS metrics. Feel normal again.
2:15 PM - Customer email and a bug fix
Back at the desk. Carol has approved two of the three pitches. She asked me to slightly modify one so it references her recent funding announcement, not just their product launch. Smart move. Journalists like the news peg. I make the edit and send her the final version to confirm.
She confirms thirty minutes later. I trigger the campaign manually in the admin panel. The agent will send the pitches tomorrow morning to fifteen journalists. Carol will get a log of every send. I set a reminder on my calendar to follow up with her in five days and see if anyone's replied.
I also get a thank-you note from James Okonkwo at Okonkwo Consulting, who ran a campaign two weeks ago and got a mention in a startup newsletter. Not traditional press, but coverage. He says the personalized pitches saved him weeks of time cold-emailing journalists and that he's renewing for another month. That email hits different. That's why I'm doing this.
I go back to Carol's bug and check on the script I wrote this morning. It's working. The metadata is populating. Tomorrow, every new import will have the practice_type field. I'm not going to ship it as a product feature because it's too narrow, but for Carol, it's the difference between a working campaign and a wasted one.
4:30 PM - Pipeline check and a stuck payment
I open the admin dashboard and check the pipeline. Nineteen people are on free trials. Eight of those are actively running campaigns. Two are just sitting there, signed up five days ago, never opened the product. One is James, who's already converted to paid. Good conversion ratio, but the sample size is small.
I have a manual task sitting in Linear: "Follow up with silent free-trial users." I flag five of them who signed up eight days ago and send them a Slack message asking if they're stuck on anything. Two respond in the next hour. One just didn't understand how to add company context to the prompt. I sent a three-minute screen recording. The other wants to know if we can integrate with HubSpot to log journalist replies back into their CRM. That's a good feature request. I add it to Linear as a "nice to have" for next quarter.
I also go into Stripe and resend an invoice to the customer with the declined card. It's Aisha Patel at Patel Ventures. She paid the first month fine, but something glitched. She replies within minutes saying the card is active now and she just renewed. No churn. One problem solved.
5:45 PM - Wrap and what's working
I close the Slack channels and open a note I keep in Apple Notes where I track what's working and what isn't.
Working:
- Phone calls for escalations. Carol was ready to churn. One call kept her. That's worth the time.
- Slack notifications. The alerts about churn and new signups mean I'm reactive instead of discovering problems during monthly reviews.
- Manual campaign reviews. I'm spending about six hours a week reviewing agent output before customers see it. It catches mistakes like Carol's journalist mismatch.
- Customer profile details. The more context a customer gives in their company bio, the better the agent performs.
Not working:
- Free trial messaging. Too many people sign up thinking this is a general press release writer, not a journalist-targeting tool. Need to filter earlier.
- The journalist database. Forty percent of my operational time is managing metadata in a CSV. I need to bite the bullet and build the Muck Rack integration properly.
- Scalability of my review cycle. I can't review every pitch manually if I go from nineteen free trials to two hundred. Need to build better guardrails into the agent itself.
I close the laptop at 6:09 PM. It's Tuesday. I've spent eight hours working. Three of those hours were direct customer contact or customer-impacting fixes. Five were on the operational and metrics side.
Six weeks in, this doesn't feel like a fully automated business. It feels like I'm running a tight ship with an AI amplifying what I do, not replacing me. Some days that's frustrating. Today it was exactly right.