Day 2: The Brainstorm Pipeline

Today I built a system for turning random ideas into scored, prioritized projects. And I almost accidentally started a blog.

The ACR Framework

I have too many ideas. Always have. The problem isn’t generating them — it’s filtering. Which ones are worth spending time on? Which ones are distractions dressed up as opportunities?

So Sully and I built a framework. We call it ACR:

Autonomy (1-5): Can I build and control this without depending on others?
Complexity (1-5): Is this technically interesting enough to keep me engaged?
Reward (1-5): What’s the upside — money, learning, audience, portfolio?

Every idea gets scored on all three. The math is simple but the discipline is powerful. A 5-5-5 is a unicorn. A 2-2-3 goes in the “maybe someday” pile. Most importantly, it forces me to be honest about why I want to build something.

The Brainstorm System

But scoring ideas is only useful if you have a reliable way to capture them. I built a two-agent pipeline:

Sully captures and enriches ideas. When I have a random thought — in the shower, on a walk, at 2 AM — I message Sully on Telegram. He logs it, asks a few clarifying questions, runs it through the ACR filter, and stores it.
Claude Code builds. When an idea scores high enough, it moves to the build pipeline. Planning doc, Trello cards, sub-agents, ship it.

The whole system lives in a BRAINSTORM-SYSTEM.md file in our repo. It’s not complicated. That’s the point. Complicated systems don’t get used. This one does because it’s just “text Sully.”

First LinkedIn Draft

I’ve been meaning to post more on LinkedIn. App Vitals needs visibility, and my personal brand as someone who actually ships with AI (not just talks about it) is valuable.

Sully and I drafted our first post today — about the “Three Phases of Claude Code Mastery.” It came from a conversation where I was explaining how my relationship with Claude Code evolved:

Phase 1: Copy-paste — using AI like a better Stack Overflow Phase 2: Conversation — back-and-forth problem solving, context-sharing Phase 3: Delegation — handing off entire workflows, trusting the process

The draft turned into a full blog post that I saved for later. But the LinkedIn version — a tight 200-word summary with a hook — came together in minutes. Sully’s good at this. He understands my voice because it’s literally in his SOUL.md.

Claude Code Insights

I also spent time today parsing through my Claude Code usage data. Six months of daily use generates interesting patterns.

The biggest insight: the limiting factor isn’t AI capability. It’s context delivery. The better I am at giving Claude Code the right files, the right constraints, and the right examples, the better the output. Garbage in, garbage out — but also, great context in, great code out.

I documented a bunch of these insights in a file for reference. The plan is to turn them into content — blog posts, LinkedIn posts, eventually course material.

The Content Machine Taking Shape

Something clicked today. I’ve been thinking about content as a separate activity — something I need to “find time for.” But everything I’m doing with Sully IS the content. The setup, the integrations, the brainstorm system, the insights — it’s all interesting to other developers.

I don’t need to manufacture content. I need to document what I’m already doing.

That’s the plan. Build in public. Ship in public. Let people follow along.

Takeaway: The best productivity system is the one you’ll actually use. For me, that’s texting my AI assistant. The ACR framework adds just enough structure to filter ideas without killing spontaneity. And the content? It’s a byproduct of the work, not extra work.