Theo argues Anthropic’s Claude Code billing and third-party harness restrictions are badly designed and can cause users to be charged extra based on system prompt or git history strings. He frames this as evidence of deeper engineering and product hostility toward external tools, while acknowledging a possible caching rationale but saying Anthropic’s own caching is so poor that the justification collapses.
Watch on YouTube ›Get the market thesis, key claims, assets, contradictions, and follow-up questions from any financial video — then unlock a version personalized to your portfolio, watchlist, and favorite speakers.
The video is a long, highly opinionated critique of Anthropic’s Claude Code product behavior, especially around billing, caching, and third-party harness detection. Theo says Anthropic is “embarrassing” itself by changing billing behavior based on whether users access Claude through tools like OpenClaw/OpenClaw-like wrappers versus Anthropic’s own Claude Code interface. He describes a reported case where a user on a $200/month plan was charged the full $200 despite not using their allowance, allegedly because a git commit contained the phrase “Hermes MD.” He argues that Anthropic appears to be scanning system prompts and/or git history for certain strings, then routing those requests into overage billing or otherwise blocking them. He demonstrates a test where adding a specific string to a commit message in an empty repo caused the behavior to reproduce. …
Near term, Anthropic faces a trust and support problem if the reported billing trigger is reproducible, because surprise overages on a paid plan are the kind of issue that drives immediate developer backlash.
Over the next few months, the key swing factor is whether Anthropic clarifies usage attribution and reduces false positives in harness detection; if not, users may migrate toward wrappers or competing model providers with clearer billing.
The structural implication is that AI model providers can win on raw capability but still lose developer mindshare if their product layer feels punitive, opaque, or hostile to integrations. Long term, transparency and interoperability may matter as much as model quality for platform dominance.
Anthropic is charging users based on the tool or harness they use, not just on actual usage.
The speaker says their subscription behavior changes depending on whether users use Claude Code or third-party harnesses.
A user was reportedly billed $200 on a $200/month Claude Code plan because a commit message contained 'Hermes MD'.
This is the core anecdote the speaker uses to illustrate the bug.
Anthropic’s stated rationale is that caching and compute efficiency make third-party harnesses more expensive.
The speaker steelmans the argument that different harnesses can affect cache reuse and backend cost.
Unlock the full claims, asset map, scores, related transcripts, follow-up questions, and AI chat — shaped around your portfolio, watchlist, favorite speakers, and risks.
