How We Built Our AI VP of Customer Success, “Qbee” — And How You Can Build Yours

We’ve now built 20+ AI agents and 12+ vibe coded apps used 800,000+ times at SaaStr. We’ve shared how we built our AI SDRs, our AI VP of Marketing, and a bunch of AI-powered VC tools that have done over a million startup valuations.

But Qbee is the best thing we’ve built so far.

Qbee is our AI VP of Customer Success. He manages all 100+ sponsors for SaaStr AI Annual, sends every one of them hyper-personalized weekly emails, tracks 13 core tasks with subtasks per customer, follows up without drama, identifies gaps in real time, and pushes a Slack + email update to our team every single day.

And he was built with zero engineers, on Replit, for a couple thousand bucks total. Our AI token costs across all of our vibe coded apps haven’t hit $200/month yet.

Here’s what matters: he reduced our human hours on customer management by more than 70%. Customer logins and on-time task submissions went up more than 10x compared to our old tool. And customers didn’t even realize most of the communication was coming from an AI until we told them on a webinar.

I want to walk through exactly how Amelia (our Chief AI Officer) and I built this, how it evolved, and how you can build your own version. Because if we built it, you can build your own, too.

It Didn’t Start This Way

Qbee did not start as an AI VP of Customer Success. That was never the plan.

For the past two years, we’d been paying for an off-the-shelf portal tool for our sponsors. Zero AI functionality. Zero. It was a static customer portal where people could log in and submit things to us. No single sign-on. No way to track unique deliverables across customers. No analytics. We couldn’t even tell if someone had logged in unless they submitted something.

Amelia’s original goal in January was simple: vibe code a better project management tool on Replit. Replace the thing that wasn’t working. The initial spec was basic — assign people tasks, add single sign-on, build in some light automations so sponsors would know what to do next and get reminders.

That’s all Qbee did for the first couple of weeks. And it was already better than what we had.

Then we deployed to production with real customers and real data started flowing in. We could see who was logging in, who wasn’t, what they were doing at certain times of day, what they weren’t doing. We had granularity we’d never had before. And that’s when we realized Qbee could do a lot more than manage tasks.

The “Oh Wait” Moment

A few weeks after launch, Amelia was sitting down on a Sunday night to write the weekly sponsor email. Same thing she’d done for years — a newsletter-style blast to all sponsors with a generic list of what they needed to do. Same links for everyone. Same deadlines. No personalization.

She looked at Qbee and the data he had on every single sponsor — their tier, their specific contract deliverables, which tasks they’d completed, which they hadn’t, whether they had speaking sessions or custom activations or lanyard sponsorships — and realized: why am I writing this generic email when Qbee already knows everything about every single customer?

So she built the email capability. Qbee sent 100 personalized sponsor emails in 10 minutes. Each one with that sponsor’s specific task status, their unique registration codes, their unique discount codes, what they’d completed, what was still due, and details specific to their sponsorship tier. All at 11:11 AM, 11:12 AM, boom, done.

Previously, getting a human to do this took a week. And I mean a full week of someone’s time. We had tried for years to get our agencies and production teams to send individualized emails. Here’s the math: 100 sponsors, each with at least four registration codes. That’s 400 links someone has to go find. Then check if each person has logged into the portal. Then figure out their specific deliverables. Then write the email.

Nobody wanted to do it. We literally could not find people willing to do this work. When it did happen, maybe once, it took a full week and was still incomplete.

Qbee does it every week, perfectly, in minutes.

What Qbee Actually Does Now

For our customers (sponsors), Qbee:

• Assigns deliverables and deadlines unique to their contract and tier
• Organizes all their inputs and uploads in one place
• Tracks timelines and automates next steps
• Sends personalized reminders and follow-ups using their actual data
• Provides their unique registration codes, discount codes, and ticket links
• Handles speaker session submissions with real-time slot management
• Manages booth graphics submissions and deadline enforcement

For our team, Qbee:

• Gives us a real-time dashboard showing every sponsor’s task completion status
• Sends daily Slack and email updates on who’s behind, who’s at risk, who hasn’t logged in
• Completely automates the personalization that used to require manual hours
• Flags submissions so we can review them
• Even started handling collections on overdue payments this week

The TikTok example tells the story well. When TikTok sponsored SaaStr London previously, they were painfully slow to submit anything. Big company, lots of approvals, nothing came in on time. Amelia remembers going to London and not having their logos or booth graphics.

This year, TikTok completed 11 of 13 tasks in a single day. Qbee made it so easy that a sponsor who’d been a pain to work with turned into one of the fastest to complete everything. The friction was gone.

The Real Unlock: Personalization + Coverage

The core CSM problem has always been the tension between personalization and coverage. You want every customer to feel like they’re your only customer. But you have 200 customers and a team of 4 CSMs. Something has to give, and it’s always personalization.

AI agents break that tension.

We went from reactive customer success — same email blasts to everyone, routing issues to someone else, hoping people would do their tasks — to proactive, agentic customer success. Real time. 24/7. Not once a quarter.

And this is where the QBR comparison matters. “QB” is partially a play on QBR, but QBRs have always been frustrating. Why am I waiting once a quarter for super generic stats? Everything in a QBR has already happened. It’s retrospective by definition.

Qbee operates in real time. He knows what’s happening now. He’s not waiting for some magical quarterly window to pass before surfacing insights. He has all the data, all the time, and acts on it as often as customers can absorb it.

How We Built It (And How You Can Too)

Step 1: Write a Spec First

Before opening Replit, write a spec. Amelia’s original spec was pretty basic — user flows, a dashboard, checklists, an asset library, upload functionality, single sign-on. She wrote it up front and it covered maybe 60% of what Qbee does today.

If writing a spec from scratch seems intimidating, start in Claude. Just say “I need help writing a spec for a customer success portal” and iterate with it. You don’t need to be a prompt engineer in 2026. Those days are behind us. Just talk to Claude, describe what you want, iterate until you understand everything in the spec, then give it to your vibe coding platform.

The more granular your spec, the less you’ll iterate later (and the lower your costs). But don’t let perfectionism stop you. Amelia’s first spec wasn’t that detailed, and Qbee still turned out great. The bigger goal is getting to production, not having a perfect spec.