8:42 AM - Inbox triage
I pour coffee and open the Slack alerts first. Three notifications are waiting: two new signups came through overnight (one in Portland, one in Denver), and our Stripe webhook just flagged a card decline on an existing customer. I open the admin dashboard and scan the numbers. We're at $2,847 in MRR this month, up from $1,920 last month, and the weekly pipeline shows $4,100 in committed demos for the next ten days. Not life-changing yet. But it's moving.
I switch to Gmail. Forty-three new messages. I filter for customer and look at subject lines. One from Carol Reyes at Reyes Family Practice asking about compliance notes integration. One from Marcus Webb at Webb Studio Architects saying thank you for the quick turnaround on his last RFI draft. One from our payment processor about the failed charge. And three marked urgent from leads who got demo links yesterday.
I start with the thank-you. Marcus sent it last night: "The change order draft you generated was spot-on. Saved me three hours. Already showing this to my partners." I save that to a wins folder in Gmail. I'll review these at month-end when I'm tempted to second-guess the whole thing.
9:16 AM - Agent review queue
I open the main product dashboard. Ten documents are waiting in the review and approve queue, the proposals, specs, and RFI drafts the AI agent generated overnight for active customers. This is my morning ritual. Each one needs human eyes before it goes to the customer. I'm not a rubber stamp. I'm the editor.
First is a project spec for Harrison Architects in Seattle. The AI pulled in the site dimensions correctly from their intake form, drafted a scope of work, and outlined the client budget tiers. I skim it. One line jumps out: "The surveyed grade elevation will be finalized by the contractor." That's vague. In real work, the contractor doesn't finalize anything the architect hasn't signed off on. I click request revision and leave a comment: "This is the architect's call. Rephrase as 'grade elevation to be verified by surveyor at pre-construction walkthrough.'" The agent will rewrite and resubmit in five minutes. This is the work. This is where the value actually happens.
The next six documents are clean. The seventh, a compliance checklist for a medical office in Portland, has a red flag. The AI included requirements for a specific ADA accessible door width that changed in 2024. I know this because I did the research last month when a customer almost slipped past code review. I update the template globally, approve this one, and make a note to check whether this needs to go to customers who've already received older versions. That's tomorrow's project. I approve the rest and move on.
10:45 AM - A customer gets stuck
Slack pings. It's Carol from Reyes Family Practice. She's a three-week customer who signed up after a LinkedIn message I sent her in March. She's been quiet, mostly used the system twice. Now she's asking: "Can I generate an RFI draft for the mechanical consultant? The current form seems to assume I'm writing to the general contractor."
This is a real problem. Our intake form is opinionated. It assumes a certain project workflow. Not every firm works that way. I respond quickly: "Yes, that's a gap. Let me generate one for you right now and we'll get this into the system for next time." I open a blank RFI template, fill in her project details manually, and generate a draft specifically for MEP consultants. I send it to her in Gmail with a note: "Try this. If it lands for you, I'll add a toggle to the form so you can choose your recipient." She'll try it. She might stay. Or she might churn in a month because we don't quite fit her workflow yet. I'm learning that some customers are early adopters. Some need us to be a little more flexible.
12:30 PM - Metrics and the hard part
Lunch is a desk salad while I pull up the analytics dashboard. I'm tracking three things obsessively right now: signup-to-first-doc conversion, currently 71 percent. Docs-generated-per-active-user, 4.2 per week, up from 3.1 last month. And churn. That last one is where my stomach tightens.
Two weeks ago, David Park at Park Collaborative Architecture sent me a message: he was canceling. He'd generated four documents, they were okay, but he found a different tool that integrates with his favorite CAD plugin. I refunded his three remaining weeks. No drama. But it was a reminder: our product has to stay valuable or these firms will leave.
I open Linear, our task management tool, and look at the backlog. Eighteen items flagged by customers in the past week: better filtering for project types, an export to Word instead of PDF, a Slack integration so they see notifications when their documents are ready. I can't build all of these. The engineering partner I work with can probably ship one or two this month. I'm learning to prioritize. I put the Word export at the top. Three of my best customers asked for it. That's the signal I'm watching.
2:15 PM - The billing mess
Stripe dashboard. The declined card I saw this morning belongs to Diane Holloway at Holloway Reeves Architects. She's been a customer for six weeks, on the $249 plan. Not a power user, but consistent. The charge declined. Could be an expired card. Could be she changed banks.
I don't automate this step. I send her a direct email: "Hi Diane, your payment didn't go through this morning, probably a card thing. Can you log in and update your payment method? Or just let me know and I can send you a new link." Two minutes later she replies: "Oh, that was my old corporate card. Updating now." Stripe shows the charge went through twenty minutes after that.
This is the unglamorous work. But it's the work that keeps someone at $249 a month instead of losing them to an inactive account. I used to think automation meant not doing this anymore. Now I think automation means doing fewer things that don't matter so you can do more things like this, the human-touch things that actually land.
3:45 PM - Lead follow-up
Three demo links went out yesterday to firms that found us through LinkedIn or the AIA chapter partnership. I check the analytics. Two of them opened the links and spent time on the site. One is from Sarah Kaufman at Kaufman Turner Associates, a 28-person firm in the Bay Area. The kind of firm we want. She spent four minutes on the page. She didn't schedule a demo, but she engaged.
I write her a personal email. Not a template. I mention that I saw she checked out the platform and ask her a specific question: "I noticed you're doing a lot of healthcare work. Are compliance notes the biggest documentation bottleneck for those projects, or is it the RFI process?" Genuine question. If she replies, I'll know what to demo. If she schedules a meeting, I'll lead with the tool that matters to her firm.
I send it and make a note to follow up in four days if I don't hear back. That's the 3-to-5-closes-per-month math. Follow-up. Specificity. Listening to what each firm actually needs.
4:55 PM - One small bug fix
A customer reported that when they generate a proposal, the client name doesn't carry over properly if it has an ampersand in it. "Smith & Jones" becomes "Smith Jones." It's small but it's sloppy. I pull up the code, find the escaping issue, and push a fix. By tomorrow morning, new proposals will render correctly. This is the kind of thing that builds trust. Not in a big gesture way. In a "we're paying attention" way.
5:45 PM - Pipeline snapshot
I look at the committed pipeline one more time before I close the laptop. Ten demos scheduled for next week. Three of them are warm, firms that are actively looking for a solution. Seven are lukewarm, LinkedIn outreach that converted to a meeting but I'm still selling the idea. Statistically, I should close two or three of those. That's another five hundred dollars in MRR, maybe more.
It's honest work. The AI isn't operating itself. I'm operating the AI. I'm reviewing, fixing, escalating, following up, deciding what to build next. Some days feel like a lot. Some days I wonder if three closures a month is sustainable when I'm already working fifty hours. But then I get a message like Marcus sent, a person whose time I actually saved, and I remember why this matters.
I close the laptop at 6:12. The coffee is cold now.