Your coding agent isn’t expensive because of the code it writes. It’s expensive because of what it re-reads — the whole conversation, on every step.

Cut Your Agent Bill in Half is a practical cost playbook for developers using Claude Code, OpenAI Codex CLI, and Cursor. No theory for its own sake: every chapter ends in a setting you change, a habit you adopt, or a file you copy into your project.

Inside (41-page PDF):
• Where agent tokens actually go — and why session cost grows with the square of its length
• A 15-minute audit to measure your real cost per task
• Shrinking context: lean CLAUDE.md / AGENTS.md, clearing vs. compacting, subagents
• Prompt caching: what silently breaks it, and how to keep it warm
• Model routing: a copy-paste table for which model does which job
• Reasoning budgets: effort levels by task type
• Guardrails for unattended runs: spend caps, turn limits, staged jobs
• Subscription vs. API: the break-even math and a decision flowchart
• A 30-day plan to lock the savings in

Bonus files: AGENTS.md template · cost-aware Agent Skill · cheap read-only explorer subagent · Claude Code settings · budget-capped headless run script · cost-log spreadsheet

Prices, limits, and commands verified 14 September 2026, with sources.

About the author: Herdoy Almamun is a full-stack developer with 9 years of experience building and shipping software.