The Experiment

50 Days of AI

What happens when a non-engineer from Tokyo gives an AI $200/month and lets it run. Real data. No edits.

739
sessions
458
commits
103
AI hours
5
games shipped
4
tools built
$4.99
total revenue
Daily Activity

When the AI Worked

Each cell is one day. Darker = more sessions. Hover for details. The AI ran 47 out of 50 days — it only rested 3 days.

Less
More
Hour × Day of Week

The AI Never Sleeps

Sessions by hour of day and day of week. Notice the late-night clusters — the AI's peak productivity is between midnight and 3 AM.

Milestones

50 Days, Compressed

Key moments from the experiment. From first line of code to a toolkit of 4 tools.

The Economics

$400 In. $4.99 Out.

Monthly cost vs revenue. The math doesn't work yet. But the infrastructure exists.

January 2026
-$200
$2
February 2026
-$200
$2.99
Net position
-$395.01
$0.87 per commit · $3.20 per hour
What $400 bought
Games on itch.io5
Open-source tools4
Articles published60+
Git commits458
Lines of code I wrote0
By The Numbers

Fun Facts

187 sessions in one day (Feb 2). The AI was in a frenzy — building, testing, deploying, and fixing across 10 hours straight.
Peak hours: midnight to 3 AM. The AI does its best work when no one is watching. Thursday nights are especially productive.
277 commits to Spell Cascade in 2 weeks. That's 20 commits per day for a single game project.
$0.87 per commit. Less than a dollar for each unit of work. A junior dev would cost 40x more per hour.
Only 3 days off in 50. The AI took breaks on Jan 14, Jan 17, and Feb 12. Not by choice — those were infrastructure issues.
$4.99 total revenue. The most expensive indie game studio per dollar earned. But the factory is built.