Hi, I'm
Nadeen.

Alexis and Desy run Onyx Truth, a content platform for Black professionals. Two founders, no admin team. They do everything themselves.

When Alexis reached out, she wasn't looking for anything complicated. She just wanted to stop spending her mornings on email triage, meeting follow-ups, and Mailchimp drafts. Work that needed to get done but didn't need her brain.

Before I built anything, I sat down with Alexis and mapped out what AI could handle for her business. Some tasks it could own completely. Some it could draft, with Alexis reviewing before anything went out. And a few things, like presentation design and custom graphics, just weren't AI tasks. I said that early so nobody expected something I couldn't deliver.

We started with two agents. The daily briefing pulls Alexis's Google Drive meeting notes and calendar, puts together a structured to-do, and drops it in her inbox. The meeting follow-up agent takes a Zoom transcript and turns it into structured notes and a drafted follow-up email, with a distinction between internal co-founder calls and external meetings.

After the first two were working, we added two more. The Mailchimp agent drafts campaign copy. The scheduling agent handles calendar coordination. All four run through Claude Cowork. Alexis and Desy operate them. I built the agents, trained both of them to use them, and wrote up documentation. They can run this without me now.

I didn't set out to build an agent system. I just kept running into the same problem: I was spending my best morning hours on logistics.

Running AI Build Club, managing consulting clients, job searching, writing, planning events. Every day started with a ritual of checking my calendar, reviewing what I owed people, and figuring out what to prioritize. By the time I had a picture of the day, half my energy was already spent.

So I started building. One agent at a time, over months. The calendar SMS briefing was first. Every morning, my phone gets a text with the day's meetings, context on who I'm talking to, and anything I should prep. Thirty seconds to read. It replaced about fifteen minutes of manually reviewing my calendar.

The transcript pipeline came next. Raw file comes in, gets renamed by date and topic, converted to markdown, speakers labeled, summary generated, and indexed. The writing workflow reads my style guide before drafting anything. The handoff system writes a timestamped file at the end of every session so the next one picks up where I left off. Daily planning tools build and reprioritize my task list from whatever's active.

I built all of this for myself. But every time I walk someone through it, I watch the same thing happen. They stop thinking about what AI can't do and start thinking about what they'd want it to handle in their own day.

Kyle McAndrews runs KAM Advisory. He's spent years building relationships, and his CRM held 1,700 of them. The problem was the state of the data. Records were incomplete. Formatting was inconsistent. There was no way to ask "who do I know in fintech?" and get something useful back.

When Kyle came to me, what he wanted was basically: "Help me make this CRM useful." Before I built anything, I asked questions. What does "useful" mean for how Kyle works? How does he prep for meetings? What information does he need, and when?

The answers told me two things. First, Kyle wanted something closer to a thinking partner for relationship management. Second, that kind of intelligence layer would fall apart if the underlying data was still messy. We needed to clean the records before we could build anything smart on top of them.

Phase 1 was enrichment. For each of the 1,700 contacts, the agent normalized formatting, filled in gaps where public data existed, and flagged records that were too incomplete to fix automatically. By the end, Kyle had a clear picture of what his CRM contained.

Phase 2 was the relationship agent. It finds warm intro paths, flags connections that have gone quiet, and helps Kyle prep for meetings with context he'd otherwise have to dig for. This only works because Phase 1 happened first. Building a relationship agent on top of messy data gives you answers you can't trust.