Video Summary - Anthropic is worse than my high school's drug dealer

Overview

The video is a highly critical rant by Lewis Rossman arguing that Anthropic’s business practices around Claude Code and related usage/billing behave like “drug dealing” or “a gaslighting ex”—i.e., they hook users with good service, then change the rules in ways that:

reduce the value users get,
increase costs,
and create workflow instability,
without clear upfront disclosure.

Key claims and examples

Claude Code is genuinely useful (in practice)

Rossman emphasizes that Claude Code has been valuable for his work, especially for tasks like:

file recovery
corrupted metadata repair

He claims open-source alternatives (e.g., “OpenAI Codex” and some open-source models) allegedly fail to produce working results.

The core complaint: billing and sudden policy changes (not model quality)

Rossman argues the issue is not model quality, but rather billing mechanics and sudden rule changes.

Examples he cites include:

A prior rule change involving a file named Hermes.md inside a session, which leads to higher “extra usage” billing.
He claims Anthropic did not clearly communicate the change before it impacted users.
After backlash, refund requests were denied (“big backlash… refunds and got denied”).
He says Anthropic later tried to address it in a different way.

“Slices of pizza” billing analogy

Rossman explains the analogy as follows:

Subscription tiers are like buying a whole pizza (cheaper bundled usage).
API/overage billing is like buying individual slices (much more expensive).

He argues Anthropic made it easier/cheaper to keep going after limits by charging extra usage at higher rates, but again—without sufficiently clear notice.

“claude-p” workflow cost shift

Rossman describes a scenario involving using Claude Code in two modes:

Inside the Claude Code program (terminal/CLI integration)
Directly invoking claude-p from a Linux terminal, enabling automation scripts

He claims:

Anthropic posted a Twitter message that sounded like users were getting a credit, but he argues it meant something else.
Using claude-p would be billed at an API-like “programmatic rate”, effectively reducing included value by ~85–90%.
Users would then need to rewrite scripts/workflows to avoid much higher costs.
Rossman says he spent an entire weekend rebuilding his infrastructure to keep costs reasonable, only to later receive an email implying another rollback/extension (“lol JK, we’re going to put that off”).

“Hooking” behavior via limits on gift accounts (test)

Rossman describes testing Anthropic gift accounts, claiming:

At first, gift accounts had a session limit but no weekly limit, unlike normal accounts—so constraints hit faster after signing up, which he interprets as intentional “getting hooked then tightening.”
Later, gift accounts allegedly had limits aligned with normal accounts—framed by Rossman as “clawing it back” closer to the IPO period.

Broader thesis: inconsistency and “hostage-taking”

Rossman argues Anthropic-style behavior is like:

unreliable relationships (hot/cold),
creating constant anxiety for people building real workflows.

He claims the root problem is inconsistency, not incremental improvement:

Businesses should provide consistency, not constantly shifting rules and surprise costs.
He draws an analogy to policy uncertainty (e.g., tariffs changing via tweets), where companies stop investing due to risk.

He concludes that even if Anthropic’s tools are strong, users may switch away because they don’t want to be “held hostage” by cryptic rule changes and hidden billing mechanics.

Overall stance

The video argues Anthropic is pursuing a strategy that:

initially delivers value,
then reduces included usage and raises effective costs,
communicates changes poorly or indirectly,
and behaves in a way Rossman considers unethical and likely to backfire over time.

Presenters or contributors

Lewis Rossman (host/presenter)

Anthropic is worse than my high school's drug dealer

Key takeaways