A founder narrated. The audience texted. AI agents built it live.

I spent two days this week at Seed the South this past week, and the most memorable moment wasn't a quote or a slide. It was my friend Charles D'Andrea standing on stage, talking out loud to Claude, while a team of AI agents built an app in front of 100 people.

He didn't show up with slides or a pre-built demo. He showed up with a microphone and Claude Code. The audience texted inputs to the agents as they worked, voted on which app to build, answered clarifying questions in small groups, and watched a live canvas update on screen as the code took shape. Five agents wrote over 1,200 lines of code in about four minutes, purchased a domain, provisioned infrastructure, and deployed it. One person and a team of agents building together in front of a live audience.

A lot of great lines landed across both days. Jonathan Donahue quoting C.S. Lewis: "Humility isn't thinking less of yourself, it's thinking of yourself less." Alan Chavez from Torta Studios: "If you don't know what your activation number is, you don't have a business."

But what made that live build hit so hard is what happened in the sessions before it. Two other presenters laid out the exact sequence that makes a moment like that meaningful, and it applies whether you're a startup or a 200-person company launching a new product.

Liz Brigham ran a product marketing masterclass that started with something refreshingly analog. She pulled someone from the audience and customer-interviewed him on the spot.

Five questions in three minutes. She asked if his problem was urgent (yes, constant). What happens if he doesn't solve it (missed relationships, dropped follow-ups). How he's solving it today (a tool that isn't built for him). And what he's spending ($10K a year on software that doesn't fit).

In three minutes, she'd validated urgency, pervasiveness, and willingness to pay. No prototype, no business case, no code. Just a conversation.

This applies to any company building anything new, whether it's a customer-facing product, an internal tool, or a process redesign. Her framework is simple and effective: get curious, talk to people, try stuff. She walked through her STOP method (Segmentation, Targeting, Observation, Positioning) and showed how to feed customer interview transcripts into Claude to build living personas you can update with every new conversation and use to pressure-test your messaging, your positioning, and your roadmap.

Her biggest point landed harder than any demo. She called every AI output your "sh*tty first draft." Use it to accelerate, not to replace your judgment. Just because it's easy doesn't mean it's right.

Alan Chavez and Colin Thornton from Torta Studios ran the session that reframed how I think about retention.

They broke down what companies like Facebook, Slack, and Pinterest figured out. Facebook found that when a user added 7 friends in 10 days, they stayed. Slack found that once a team hit 2,000 messages, they'd replaced email and weren't going anywhere. Pinterest found that 3 pins in the first week gave the algorithm enough signal to personalize the feed and keep people coming back.

Every product and service has its version of this number, and most companies never look for it. That includes your customer portal, your employee onboarding, your SaaS tools, and your internal platforms.

The framework is pretty straightforward. First define your retained users, then watch what they do differently from the ones who churn, and isolate the specific events that predict retention. The insight isn't that people leave because something broke. It's that retention is usually decided in the first session, sometimes in the first few minutes.

Here's where it gets interesting for AI. They ran a live demo where an AI analytics tool analyzed real product data and confidently recommended "dashboard shared" as the activation event, but it was wrong. The actual event was "teammate invited." You can't share a dashboard unless you already have someone to share it with. That's reverse causation, and the AI missed it completely.

The tools are incredible at running the queries and building the charts. They can't tell you what actually causes retention, that part is still yours.

And that brings us back to Charles's session.

What made it powerful wasn't just the spectacle, it was the process. Before a single line of code was written, the audience decided what to build. Claude texted everyone in the room, collected their biggest challenges, proposed three concepts, and let them vote. Then Claude texted clarifying questions to small groups to refine the design.

By the time the five agents started building in parallel (three on frontend, two on backend), the problem was validated and the spec was shaped by real human input. The agents purchased a domain for $7.50, provisioned the infrastructure on Cloudflare, and deployed it live.

Here's why this matters for mid-market companies. The cost and timeline of building solutions has fundamentally changed. What used to require a six-month project, a full dev team, and a six-figure budget can now be prototyped and tested in hours. That changes how you evaluate build vs. buy. It changes how fast you can test ideas with customers before committing resources. And it changes what's possible for teams that don't have massive engineering departments.

Charles made a point that applies to any leader evaluating AI. He spends hours planning and designing with Claude before writing any code. The thinking is where humans add value. His advice is to "ride the exponential." Don't over-invest in solutions for problems the model will solve next month. If something works 95% of the time today, ship it and let the model catch up.

These takeaways apply to every company, regardless of size or stage. Talk to your customers, find your number, then build it. The tools are ready.. the question is whether you're pointed at the right problem.

Putting this into action

What I’m reading

Some AI tips for this week

I’ve got 8 more tips in my recent blog post on the StealthX website.

Onward & upward 🤘
Drew