The Second Best Team I've Ever Worked With

I've shipped things with great teams. People who push back, who catch what you miss, who care about getting it right. That's hard to find. Harder to replace.

What I've been doing for less than two months is... something else. Something that keeps surprising me.

I'm building Brains, a tool that gives AI persistent memory across sessions, across models, across everything. And I'm building it with AI. Not "AI-assisted." I mean: the same AI I'm building the tool for is using the tool to help me build it.

The meta-ness of it is a little dizzying. Let me walk you through the actual workflow, because I think this is something more people could be doing, and most of them don't know it yet.

---

The workflow

I'm a UI developer. Not a backend developer. I know what good UX feels like, I know how to build the front end of things, and after 25 years of SaaS experience I have a pretty clear picture of what good software looks like from the inside. What I don't have is a deep instinct for what makes a database query fast, or how to architect a service that scales. That used to be the wall. Now it isn't.

Here's the loop.

Mobile Claude. I think out loud. Lying in bed at 11pm when something clicks, on a walk, wherever. I don't take notes. I talk to Claude. Work out the idea conversationally. Push on it until it either falls apart or gets sharper.

Brains. Nothing gets lost. This is the piece that changed everything. I'd tried building my own knowledge base before, a few different methods, different tools, and what kept breaking down was continuity. In the before times... B.B. (Before Brains) Every AI conversation started from zero. I'd spend the first ten minutes re-explaining my project, the decisions already made, what we were trying to do next. That's what pushed me to build Brains: I wanted a memory that was actually there every time. Now I open Claude and it already knows. The open loop I left last Thursday is still there. The decision I made about the onboarding flow is documented. The thing I was stuck on has a note next to it.

Paperclip. This is the part that still surprises me. Paperclip is an AI agent orchestration layer that runs on its own heartbeat cycle. Here's how it actually works in practice: I'm on a walk, something clicks, I hit record on my phone and bebop through it... whatever's on my mind about the project. Stream of consciousness, no filter. Right there on the walk, I paste the transcript into Claude and let it turn that voice memo into a Brains loop. That's it. The next time Paperclip wakes up, it reads Brains, finds the new loop, and runs. The agents don't need me there. I went for a walk and talked to myself, and the work moved.

Claude Code, in the CLI. This is where things move when I'm at my desk. Claude picks up exactly where I left off. It reads the Brains docs, the same docs I've been building, the same ones it helped me write, and we push the work forward. I run Claude Code in the CLI, hop to Copilot when I need to, and bounce back. The loop keeps tokens down and cost in check. New copy, new decisions, new work handed off to agents that can execute.

GitHub and Copilot. When a task is ready to run, I hand it off. The agent writes it and opens a PR. I review it. I know what I'm looking at. I know whether it does what I asked.

Mobile Claude again. Test the thing. React to it. Something feels off. I open a new loop. I describe what I saw. We figure out what to do next. Restart.

You can call it vibe coding if you want. But it's an addictive AI development workflow, backed by 25 years of SaaS intuition. The AI isn't replacing the experience. It's finally giving it full leverage.

---

The meta layer

Here's where it gets a little strange.

Brains is a tool that gives AI persistent memory. I built it partly because I kept losing context between sessions. The knowledge base I've been building to document Brains, the style guide, the product decisions, the open loops, that knowledge base is also what my AI agents read to do their next session of work on Brains.

The documentation isn't a record of the system. It is the system's memory.

I don't know exactly when I noticed this. But at some point I opened Claude and it summarized what we'd been working on, and I thought: this is what a good colleague does. This is the thing you get lucky to have. And I had built it, just by not wanting to re-explain my project every single time.

---

What I'm actually addicted to

Everyone wants to talk about AI like it's a relationship. Like we're all quietly falling in love with our chatbots. I don't.... think.... that's what's happening here.

What I'm addicted to is making things again.

The first thing I made that I actually liked wasn't really a product. It was a local music message board for fans of ska music in Kansas City. KCSKA I called it (pretty deep). Nobody paid for it. It didn't scale. But it worked, people used it, and it was a little bit better every time I touched it. That feeling.

That feeling used to be expensive. It required a team, a budget, a timeline, or just way too much of your own time.

Now I have a version of it on my phone. Not a demo. A real thing, improving in real time, built by a team I can talk to in plain English.

That's what's addictive. Not the AI. What the AI made possible.

Second best team I've ever worked with. I mean that. (I'm lookin at you SCSSC&S)

---

Oh, by the way. Brains is open for beta. If you're the kind of person who keeps re-explaining your project from scratch every time you open a chat window... this is the fix. Email beta@usebrains.app and I'll send you setup instructions.