Drip
Blog · AI · Cost optimisation

How AI agents bleed your tokens.

Analysis, benchmarks, and practical guides on getting more out of Claude Code, Codex CLI and Gemini CLI without blowing through your monthly budget.

AI cost optimization· Claude Code

Why your AI agent re-reads the same file 30 times a session — and what it costs you

Every modern AI coding agent burns 34–88 % of its input budget re-sending files it already knows. Here's why the loop happens, what it adds up to in dollars, and the cheapest fix.

6 min read
Benchmarks· benchmarks

The hidden cost of re-reading: a benchmark across 8 languages

We measured DRIP on production-shape codebases in 8 languages. Here's the savings breakdown, the methodology, and what surprised us about how compression behaves on Java vs Python.

7 min read
Practical guides· Claude Code

Building token-efficient AI workflows: a practical guide

Concrete patterns for getting more out of Claude Code, Codex CLI and Gemini CLI without burning your monthly budget by lunchtime. Includes config snippets, hook recipes and a checklist.

8 min read