Authority Hacker Podcast – AI & Automation for Small biz & Marketers Your AI Bill Doubles Next Month
Mar 19, 2026
They warn that hidden double-usage promos are ending and AI bills may spike soon. They unpack huge context windows, token trade-offs, and when bigger context can hurt quality. New cheaper subagent models like GPT-5.4 Mini and Nano are explored for cost savings. Tools like Cowork Dispatch, Ahrefs’ API and Firehose, and a leaked pivot toward enterprise and coding are discussed.
AI Snips
Chapters
Transcript
Episode notes
New Model Honeymoon Often Followed By Quiet Quality Drops
- Model distillation and quantization can make newer high-context variants cheaper but slightly lower quality over time.
- Gael and Mark discuss the "honeymoon" where a new model feels great then is subtly diluted in production.
Delegate Routine Tasks To Mini Models To Cut Costs
- Use smaller distilled models (Mini/Nano) as subagents for high-volume or simple tasks to cut token costs dramatically.
- Gael suggests delegating summarization and file-reading to mini models and keeping the expensive main model for decisions.
Controlling Claude From Your Phone With CoWork Dispatch
- CoWork Dispatch turns your phone into a remote controller for Claude, mirroring the desktop chat and accessing files and browser actions.
- Gael says it synced his chat and read his email, requiring the computer to stay on but easy to set up.