Shipping on a 10M-Token Budget

Description and use

Managing AI token usage for zero-cost development using test-first generation and a token budget.

Problem

Managing AI token usage to keep monthly costs zero while developing with a free tier.

Outcome

Zero-dollar month achieved by disciplined test-first generation, token budgeting, and minimal manual edits.

Workflow

• Write a failing test for the new feature
• Send the test to the model and request the smallest implementation that passes
• Record the prompt and completion token counts with token-budget.sh
• Check remaining allowance before each generation
• Verify the output by running the test and reviewing the diff
• Merge only if the test passes and the diff is clean
• Deploy by pushing to GitHub and hitting the health endpoint
• Repeat for each subsequent feature or refactor

Evidence

The monthly bill was zero dollars.

Source

Write the failing test before the model writes the code.

Source

Three gates stand before every generation.

Source