What I built after Darby's bootcamp · Gen AI University (GenAIU)

What I built after Darby's bootcamp

Hey everyone,

Sharing this before today's session because I had shared it with my mastermind but I thought it might be useful for the others to see what's possible after going through Darby's material once already.

I want to be clear - I am not a developer but I am an operator that has been hiring, training, managing and putting teams that execute for many years.

I went through the first class, sat with the folder structure Darby walked us through, his Chief of Staff and then added a few more pieces that I've learned and spent Saturday this weekend building this out.

Wanted to show you where I landed so you can see what's on the other side of the work and sometimes seeing an example of a real work flow can help you see how you can put your own pieces together.

Two things I want to set up before I get into the team, because they're the real lessons that I think are mission critical:

Context and clarity are everything. This works because I gave each agent a clear role, a clear set of inputs, a clear definition of what good looks like, and clear hand-offs to other agents. Vague prompts produce vague work. Same as humans. If you can't write a clean "Job To Be Done" description for a person, you can't write one for an AI.

I'm starting with my first team on low-risk work. On purpose. Same way I'd onboard a new human hire.

Start them on work where the cost of getting it wrong is small. Watch how they perform. Build my own skill at managing them. Build trust. Then expand the scope. As I get better at managing the team and they get better at delivering, I add responsibility, just like a real team member. The mistake most founders make is handing AI a high-stakes task on day one and then deciding "AI doesn't work" when it underperforms. You wouldn't do that to a new hire.

Here's what I built.

A 7-agent AI team that reports to me each morning. I named them after Harry Potter characters because I'm bad with names and faces, and I started with ONE Mrs. Weasley as my Head of HR, and then she actually started naming the rest based on giving each one a personality that matched their role. It's been fun.

Quick credit before I go further: this is a mash-up of three things. Darby's folder structure & Obsidian memory layer, which we built into Thelma my Chief of Staff. Then I added the HR framework, which is how I knew what to build into Mrs. Weasley. And my own operations brain for what I would actually hire, onboard, and manage if I were building this team out of humans. The AI didn't invent the org chart. I gave it the org chart. That's the whole point.

The team

Thelma — Chief of Staff. Master orchestrator. Routes work to the team. Strategic decisions stay with me.

Mrs. Weasley — HR Manager. Owns the roster, daily check-ins, weekly team report. Diagnoses what breaks.

Hermione — Time Buy-Back. Morning brief, inbox triage, calendar defense, meeting follow-ups. The daily admin layer.

Harry — Revenue Generation. Pipeline, LinkedIn, Substack, outreach, proposals, buyer research.

Ron — Retention & LTV. Client onboarding, case studies, referrals, partner cadence, at-risk flags.

Luna — Content Intelligence. Daily niche scan, weekly content matrix, competitive watch. Feeds Harry and Ginny.

Ginny — Content Production. Daily LinkedIn drafts, Wednesday newsletter, Friday repurposing pass.

How this morning, Monday morning, went when they all fired off before I sat down at my desk.

I sat down at my computer. Coffee in hand. Waiting for me were briefs from each agent. Hermione had already pulled my morning brief together. She read my memory file (this is how the system remembers context across sessions), scanned my inbox, checked my calendar, looked at my pending folder, and surfaced what I committed to yesterday so I could see it next to today's plan.

Harry and Ron were quiet because the data feeds aren't fully wired in yet. That's next.

Luna delivered her daily intel report on the topics I'm tracking for content.

Ginny took Luna's intel and came back with content angles I could actually use today.

While reviewing my Slack I noticed that Luna failed to post her checkin but I had it in my Claude thread so I pinged Mrs. Weasley and told her that Luna didn't post to Slack and then I went and refreshed my coffee.

Then Mrs. Weasley did her check-in. She told me everyone fired except Luna, whose Slack integration had a hiccup. By the time I got back, Mrs. Weasley had diagnosed the problem and told me exactly what I needed to do to fix it.

That's the part I want you to see. I didn't write code. I didn't troubleshoot. I told her what was broken, walked away for two minutes, and came back to a fix. That's the difference between AI as a chatbot and AI as a team.

A few things to take from this:

Memory is the unlock. Every agent reads from a shared memory layer. End of day I tell them what we got done and what's outstanding, and the next morning they pick up where we left off. This is what most people miss when they get frustrated with ChatGPT forgetting them. You have to build the memory.

File management is the second unlock, and nobody talks about it. Every agent knows where their work goes. When Ginny drafts a LinkedIn post, she saves it to 06-Content/drafts/2026-05-04-linkedin.md. When Mrs. Weasley writes the weekly report, it lands in 07-Operations/team/weekly-reviews/. When Luna delivers daily intel, it has a home. File management is the administrative headache that kills most "I'll just use ChatGPT" attempts. The work gets done, then it gets lost. My agents know the folder structure Darby walked us through, and they put their work where I can find it tomorrow. That alone is worth the build.

Each agent has a scoring rubric for their work that I created off my brain and what outcome I wanted. Ginny doesn't just write a LinkedIn post and call it done. She runs it through a post-scorer skill I built, scores it against my voice and my guidelines, and any draft below an 8 gets reworked before it comes to me or saves. Same idea across the team. They know what good looks like before they create the thing, not after. That's how you stop getting AI slop and start getting work you'd actually publish.

Each agent has one job. Hermione doesn't try to do sales. Harry doesn't try to write content. Luna researches but doesn't draft. Ginny drafts but doesn't research. Specialization beats generalist every time, even with AI.

The chief of staff and HR layer are what make it run. Thelma routes the work. Mrs. Weasley watches the team and flags when something breaks. Without those two, you have 5 agents doing things and nobody watching the system.

This took a weekend. Not a year. Not a $50K consultant. A weekend, because I already knew what the operations needed to do. The AI piece was just plugging in the right key parts of the puzzle and tools to handle the steps I already had mapped.

That last point is the one I want you to sit with. The reason this worked is because I came in with operational thinking already done. I knew what I wanted my morning brief to contain. I knew the difference between content intelligence and content production. I knew where my pipeline lived. I knew what good content looked like before I asked an agent to make it. I knew where every file should land.

The AI didn't figure that out for me. It executed the structure I gave it. I did the pre-work to set it up for success, again, like you do with a human team member.

And I'm starting small on purpose. Low-risk work first. Build my own skill at managing them. Add responsibility as trust gets earned. Just like a real team.

If you're going through Darby's class right now and feeling like the folder structure piece is "the boring part," I want to flag that it's actually the load-bearing part. Stick with it. The team can't function without it.

Here's to building what's next!

Kristen

7 comments

What I built after Darby's bootcamp