Crashing out at Anthropic and getting Pi pilled

· video · Source ↗

Summary based on the YouTube transcript and episode description. Prompt input used 79979 of 95234 transcript characters.

Theo and Ben dissect Anthropic’s bad week: source code leak, 8,100 wrongful DMCA strikes, rate limit cuts, and why minimal coding agent Pi is beating Claude Code on token efficiency.

  • Anthropic sent a DMCA report to GitHub that resulted in 8,100 invalid takedowns versus ~150 valid ones — an 80x false-positive rate.
  • Theo received a personal DMCA strike for a one-line PR to Anthropic’s public (non-source) Claude Code repo, later reversed.
  • Anthropic’s Claude Code subscription subsidizes inference at 20x or more, a practice Theo argues was launched solely to lure back founders Boris and Kat from Cursor.
  • Anthropic is the only major lab that charges for cache writes: $6.25 per million tokens versus OpenAI’s $0.
  • Rate limit cuts were announced at 12:46 PM Pacific — two hours after the peak restriction window had already ended, leaving users with no warning.
  • Pi coding agent uses only 4 tools and a ~20-line system prompt; Claude Code sends tens of thousands of tokens of overhead per request.
  • OpenAI gives free Codex inference to open source developers; Anthropic began doing the same only after Theo publicly pressured them, per his account.

2026-04-09 · Watch on YouTube