Same Claude, twice the tokens: summarize a call, then re-analyze its own summary
JenniferHli · x · 2026-07-28
A small but very real token-eating loop:
- The author’s colleagues use Claude to summarize customer calls into a doc.
- Then they paste that same doc back into Claude to expand and analyze it again.
- The punchline: even when it’s the same model, it can end up generating work for its own earlier output.
It’s a neat example of how AI workflows can accidentally become recursive and expensive.
More from Fun
- Bittensor SN95 post turns “Actual Kimi” into a running AI joke — markjeffrey · 2026-07-28
- EmblemVault pitches no-code onchain apps, and the screenshot turns into a meme — adamamcbride · 2026-07-28
- Agent Arcade: An Open-Source Tool to Play Retro Games While AI Codes — DanWahlin · 2026-07-28
- Meme claims reading 2.8 trillion parameters would take a human about 30,000 years — airesearch12 · 2026-07-28
- “You don’t need the latest model” becomes the new CRUD-app reality check — georgemillo · 2026-07-28
- At least someone who only talks football can be tuned out — NathanpmYoung · 2026-07-28