Build with AI: From Zero to Your First App · Module 6: Working reliably with AI coding tools
Quota Discipline, No Full Rewrites
Quotas are the physics of AI work: every plan has them, every generation spends them. This lesson spends them like money and bans the most wasteful purchase of all.
9 min reading
Objectives
- Explain quotas and why they apply on paid plans too
- Prefer targeted edits over full rewrites with reasons
- Plan a week of AI use inside a free-tier quota
- Recognize rewrite pressure and answer it with a smaller step
Why this matters
Beginners burn quotas the way beginners burn time: full rewrites for one-line faults, five regenerations hoping one is perfect, parallel chats exploring the same dead end. A free tier covers course-sized work comfortably when spent on single changes; it starves in days when spent on rewrites. Discipline here is budgeting, not miserliness.
Concepts
Quotas recap (M01L3). How much you may use per day or month: messages, generations, tool calls. Paying raises the ceiling, rarely removes it. Every course lab states its expected message count; count yours for a week and compare. Awareness alone halves usage.
Edits beat rewrites. A targeted edit costs one generation and preserves verified work. A full rewrite costs many, discards every passed check, and reintroduces old faults in new costumes. Request rewrites only when the structure itself is wrong (wrong layers, wrong scope); fix faults with edits. "It is easier to regenerate" is true once and expensive every time after.
The weekly plan. Batch questions (three faults, one chat, summary attached), prefer follow-ups over restarts, snapshot before experiments so failures cost a restore rather than a rebuild conversation. When the quota runs low, switch to paper: briefs, criteria and rename tables need no generation at all.
Worked example
Fault: price misaligned at 360px. Rewrite path: regenerate the whole page (five generations, new faults in the form, two evenings). Edit path: one pointed message with screenshot description (one generation, one rule changed, verified in a minute). Same fault, 5x cost difference, opposite reliability. Multiply by a course of thirty faults and the discipline decides whether the free tier suffices.
Expected result: a one-week AI-use plan inside free-tier limits with rewrite pressure answered by a smaller step, written down.
The common wrong move
Regenerating until perfect: five full generations hoping the tool nails it. Perfection by lottery costs quotas and teaches nothing. One edit, one check, iterate: skill compounds, lottery does not.
Lab and next step
Lab A18 (paired with lesson 23) includes the quota log: every message counted for the move. Next, module M07 systematizes debugging: reproduce, report, test, and escape the repeated-fix loop.
Quick check
An optional 4-question self-check. Answers never leave your device, are not stored, and never count toward any assessment.
Lesson feedback
No published feedback yet.
Log in and complete the lesson to leave feedback.
Exercise
Log every AI message for three days with its purpose (edit, question, rewrite). Count rewrites versus edits; write the one rule that will halve your rewrites next week.
Pass criteria
Three-day log with purpose per message; rewrite/edit counts computed; one halving rule written (e.g. screenshot-pointed edits only, no full regenerations).