Authority Hacker Podcast – AI & Automation for Small biz & Marketers Your Claude Code Tokens Are Disappearing
40 snips
Apr 3, 2026 They unpack a massive Claude Code leak that exposed system prompts, hidden features, and a new Mythos model tier. They flag a sudden token limit cut and share ways to shrink token use. They demo EmDash, Cloudflare’s serverless WordPress alternative. They cover OpenAI’s self-serve ads roll out and Gemini’s one-click ChatGPT import move.
AI Snips
Chapters
Transcript
Episode notes
Use Lower Tier Models For Execution Tasks
- Assign cheaper models per skill: use Sonnet or Haiku for execution and simple reads, Opus only where high reasoning is required.
- Gael highlights a hidden slash command to use Opus for planning and Sonnet for execution to save tokens.
Trim Chat Histories And Cloud.md To Cut Tokens
- Keep chats and cloud.md lean; long chat histories and heavy cloud.md increase token costs because they're sent with each message.
- Gael recounts refactoring a 700-line skill.md down to 60 lines to drastically cut tokens.
Replace Token Heavy Workflows With Scripts
- Build local scripts (Python) for repeated tasks so the model calls fewer tokens and runs logic off-chain.
- Gael says many skills should use scripts to automate work without burning large token volumes.