One Month Into the 10K Commit Challenge

The Scoreboard

I started the 10K Commit Challenge on my birthday, March 21, 2026. The deal I made with myself: 10,000 commits in 365 days, AI-augmented, built in public. That's ~25 commits per day, every day, for a year.

Thirty days in, here's where I sit:

Faster than plan. But the number hides the shape, and the shape is what's interesting.

The Shape of Month One

The curve is not linear. I spent the first seven days on a family trip and barely touched a keyboard. At the end of week 1 I was at 36 commits against a week-one goal of 200 — 18%. If I had been optimizing for a clean-looking chart, I would have felt cooked. I wasn't. The challenge is 365 days, not seven.

What I didn't expect: the bounce-back was disproportionate. Week 2 more than made up for week 1, and weeks 3–4 kept accelerating. That acceleration wasn't willpower. It was the workflow compounding — I was learning how to run Claude Code across multiple projects in parallel, how to batch work into tight loops, how to let the agent handle the 80% while I reviewed and decided.

One day in week 4 I shipped 228 commits. That was the day everything clicked.

What I Actually Built

Commits are just a proxy. The real question is: what exists in the world now that didn't 30 days ago? This month produced real progress across a handful of projects:

• Several Roblox games that bridge to other engines — exploring some potentially useful technology while I have time to do it.
• Personal site & 10KCommit tracker — the scoreboard you're reading this on, open-sourced for anyone who wants to fork it.
• GitPulse — a menu-bar commit tracker for macOS and Windows.
• Tax prep & productivity glue — boring, necessary, counts.
• ICM trading system — quantitative trading infrastructure, the seed of a longer-term bet.

The projects aren't related. That's the point. A pre-AI workflow forces you to pick one and commit for weeks. An AI-augmented workflow lets you hold many in your head and rotate.

Seven Learnings From Month One

1. Starting slow doesn't kill you. Not starting does.

Week 1 was 18% of pace. It didn't matter. What mattered was that I shipped something on most days (even the vacation ones) so there was no psychological "streak broken" moment. If you're planning your own challenge, do not wait for a clean start date. Start from wherever you are, even if it's bad.

2. The ramp is a feature, not a bug.

Weeks 1→2→3→4 went 36 → 314 → 425 → 701. I wasn't working 20x harder by week 4. I was working with 20x less friction. You will not hit your target on day 1. You will learn to hit it. Budget for the learning curve instead of resenting it.

3. Parallelism is where the multiplier lives.

Single-project AI-augmented dev is faster. Multi-project AI-augmented dev is a different category. I run Claude Code in multiple terminals, each with full context on a different project. While one is running tests, I'm reviewing another's diff. Cycle time per project drops, and wall-clock output per day goes up without extra hours.

4. Commit size genuinely doesn't matter — but commit cadence does.

Some of my commits are one-line typo fixes. Some are 2,000-line refactors. Both count. What actually matters is: are you stopping at natural checkpoints and committing? The act of committing forces a "does this make sense?" pause that catches 80% of AI-generated nonsense before it compounds.

5. Public tracking changes your default.

I didn't believe this before I built the tracker. Now I do. When the chart is live and anyone can load it, you ship on the days you don't feel like it. The scoreboard doesn't care about your mood. That's the whole point.

6. Tool mastery beats tool selection.

I've tried most of them. The productivity delta between Claude Code, Cursor, Copilot is real but smaller than the delta between using any of them lazily and using one of them well. Pick one, go deep, get to the point where your keystrokes per decision is minimized. Worry about the comparison later.

7. Burnout is a real risk and you have to design against it.

700 commits a week is unsustainable if every week is a max-effort week. I'm already noticing that my best days are ones where I start with a clear queue of work, not ones where I'm context-switching from reactive stuff. Next month I'm going to try time-blocking mornings for deep work and keeping the reactive/comms work after lunch. If you're planning your own run, plan rest days explicitly, or your body will plan them for you.

A Field Guide: If You Want to Do This

If you're thinking about starting your own challenge — at any tier, Explorer (2,500), Builder (5,000), or full 10K — here's the shortest path I can give you:

Before you start

• Pick the tier that scares you a little. Too easy and you won't change your behavior. Too hard and you'll quit in week 2.
• Pick a start date and announce it. The announcement is 80% of why it works.
• Fork the tracker. Don't waste a week building dashboards. Use mine, point it at your GitHub, deploy it free on Cloudflare Pages.
• Line up 3–5 real projects. Not one. You need rotation when you're stuck, and AI tooling makes rotation cheap. Mix serious and playful — you'll end up with one sleeper hit you didn't plan for.

Week one

• Commit every day, even if it's small. Break the "I'll start Monday" instinct early.
• Don't over-engineer your workflow. Ship first, optimize later. You'll know what to tune after week 2.
• Turn on a daily commit tracker in your menu bar. (GitPulse or similar.) Seeing the number in your face every hour is surprisingly motivating.

Weeks 2–4

• Start running multiple AI sessions in parallel across different repos. This is the unlock. One session in a terminal tab per project.
• Commit at every natural pause. Between "agent finished a thing" and "I'm about to ask it to do the next thing" is the right moment.
• Review every diff. The challenge is worthless if you're shipping slop. The 15-second diff glance is your quality gate.
• Share a weekly update with #10KCommit on whatever platform you care about. Even one post. The act of writing it forces reflection.

When you hit a bad week

• You will. I did in week 1. Don't recalibrate the target. Recalibrate the next seven days. The goal is 365 days, not any one of them.
• Don't try to "catch up" in a single heroic weekend. That's how people quit. Just get back to the daily rhythm; the chart averages out.

What to avoid

• Don't game it. Empty commits, rename-spam, chaff-on-the-main-branch — fine technically, but you'll know, and the whole point is to prove something to yourself.
• Don't pick only one project. You will plateau. I guarantee it.
• Don't skip the weekly share. It feels optional. It isn't.

What I'm Changing for Month Two

Based on what worked and didn't:

• Explicit recharge time. Not a fixed schedule — no "one day off per week" commitment — but a commitment to protect recharge time when I need it. The body knows, and the commits average out over 365 days either way.
• Deeper Roblox focus. The cross-engine Roblox work is where the most real-world upside lives, and I was spreading too thin. Rotation is good; dilution is not.
• Monthly posts, daily tracking. The tracker keeps updating in near-real time, but from here on I'll write these long-form recaps monthly instead of weekly. Weekly posts started to feel thin — a month is enough time for a real story to show up.
• Recruit a few more people. The challenge is more fun with company. If you're doing this, let me know — I'll track your run on the site.

The Bigger Picture

The interesting thing about month one isn't the number. It's that the number was achievable in the first place. Five years ago, 1,684 thoughtful commits in 30 days across this many projects would have required a team. Today one person with good tools and a decent tracker can do it.

That asymmetry isn't going away. It's widening. The 10K Commit Challenge is just a way of forcing myself — and hopefully you — to live inside that asymmetry instead of reading about it.

Eleven months to go. See you next month.

Follow along: daily tracking at brentvincent.com, monthly write-ups here · Fork the tracker: github.com/brentvincent/10k-commit-challenge · Hashtag: #10KCommit