Sometime in early May 2026, the combined commit count across my three production repositories crossed 1,000.
- matt-agent: 659 commits in 13 days. The marketing automation system with 14+ AI agents.
- wimper-ops-agent: 291 commits in 25 days. Operations platform with 17-milestone pipeline and partner portals.
- medicare-agent: 90 commits in 28 days. Medicare analysis tool with RAG, plan comparison, and document generation.
One developer. Zero employees. All built with Claude Code and Gemini CLI.
How
Each commit is small. A feature, a fix, a configuration change. Describe what you want to the AI coding tool. It writes the code. You test it. You commit. You move on.
The velocity isn’t about typing speed or technical skill. It’s about knowing what you want to build. I’ve spent 10 years in financial services. I know the workflows. I know what data matters. I know where the bottlenecks are. The AI handles the code. I handle the decisions.
The tools
Claude Code is the primary development environment. Not VS Code with Copilot. Not ChatGPT. Claude Code sessions in the terminal where I describe what I want and build it conversationally. Every agent, every skill, every dashboard feature in the system was built this way.
Gemini CLI is the second tool. Added it for capacity. When Claude usage is high, I switch to Gemini. Same workflow, different provider. The strategy is using multiple AI providers to avoid hitting rate limits on any single one. More building time per day.
What 1,040 commits looks like
Three production applications. Each one is a full system, not a prototype.
The marketing system has a dashboard with 10+ apps, an event bus, cron scheduling for 30+ jobs, email warm-up scheduling, social media automation, prospect qualification scoring, content editorial workflows, and health monitoring that fixes its own problems.
The ops platform tracks real client implementations through a 17-milestone pipeline with role-based dashboards, document auto-detection, and partner portals.
The Medicare tool does conversational research over CMS data, compares plans, models drug costs, and generates enrollment documents.
All three share the same architecture pattern: SQLite for data, Docker for isolation, Claude Code for development.
What it doesn’t mean
1,040 commits doesn’t mean the code is good. It means the systems work. There’s a difference. A career engineer would look at parts of this codebase and cringe. That’s fine. The agents run. The dashboard loads. The emails send. The pipelines track.
Good enough to work is good enough when you’re a solo operator building infrastructure for your own businesses. Refactoring can happen later. Shipping matters now.
The uncomfortable part
All this velocity produced $0 in new revenue from the marketing system. The ops platform serves real clients, the Medicare practice has real income, but the marketing automation, the biggest and most complex of the three, is pre-revenue.
1,040 commits and zero marketing dollars. That’s the honest snapshot.
The bet is that this infrastructure pays off over months, not days. The build phase is done. The market response phase just started.