1 / 5
44,785
messages to an AI in 31 days

What I built, what it cost,
and what I learned.

One person. One month. Claude Code.

Feb 26 – Mar 28, 2026  ·  Claude Code
31 days produced

What I Built

stonerOS
Personal AI Memory System
22 agents, 30+ slash commands, autonomous scheduling, hooks, and a full skill layer. My entire thinking infrastructure, externalized.
Swift / SwiftUI
3 Native macOS Apps
Drift, Iceberg, and ClaudeUsage—designed and shipped from scratch. Full HIG-compliant UI, data persistence, and App Store targets.
Portfolio
Public Portfolio Site
7 redacted case studies, an AI Brain writing series, a brand system, and a deployed site—from zero to live at josh-stoner.github.io.
Automation
Full Automation Layer
7 scheduled autonomous tasks, Safari bridge, Playwright integration, iPhone data pipeline, and a launchd-based cron system.
Cost & ROI

The Math

$100
Max subscription / mo
$5,521
API-equivalent value generated
53×
return on investment
Based on Anthropic API pricing: Opus $15/$75 per million tokens input/output
Activity heatmap

When the Building Happens

7am EDT Morning peak ↑ 3pm dip Afternoon surge ↑ 10pm peak ↑ 3am
Morning peak (11am–2pm)
Afternoon surge (4–7pm)
Evening flow (peak 10pm)
3,631 messages at 10pm — single busiest hour
What I learned

Three Things That Stuck

The tool disappears
After week one, I stopped thinking about Claude Code as a tool. It became the way I think through problems. That shift—from "using software" to "thinking with a collaborator"—is the thing I didn't expect.
Speed changes what you attempt
When building takes hours instead of weeks, you try things you'd never greenlight. Three of my projects started as "let me just see if..." That sentence used to mean nothing. Now it means something ships.
The leverage is real
One person, one month, an entire system. The gap between "I could build that" and "I built that" has never been smaller. That's not a productivity tip—it's a structural change in what one person can do.