Series: Solo + AI Delivery Efficiency — Part 3 of 6.

The cost inversion

My cost-per-commit went down as I shipped more. That's backwards from how software usually works — and it's the whole point.

On a metered, pay-per-token setup, more output means a bigger bill. Linear, sometimes worse.

On a flat subscription, the maths inverts:

• May, single agent: ~$0.27 per commit
• June, parallel agents + workflow: ~$0.074 per commit

Same $100/month. Roughly 3.6× more output for the same spend, so unit cost fell ~70%. Marginal cost of the next commit: effectively zero.

And this isn't just my anecdote — it's structural. The Max 5x plan is a flat $100/month; community analysis pegs the API-equivalent cost of heavy, fully-utilised usage at around $15,000/year. The flat rate rewards heavy, sustained use; metered API billing rewards the opposite. If you're running agents hard, the subscription model quietly inverts your cost curve.

For a solo founder in an emerging market, that's the part that changes what's actually buildable. The scarce input stops being "AI compute" and becomes "my own focused time."

Sources

Figures are paraphrased; verify against the originals before quoting.

• Max plan official overview (flat subscription, Claude Code included): support.claude.com
• API-equivalent value analysis of heavy usage (~$15k/year): codeculture.store

Previous: Part 2 — Two Levers, Two Jumps · Next: Part 4 — What a Team Would Have Billed