Detailed Analysis
Claude Code's `/usage` command, which displays bar graph visualizations of a user's current timeslot consumption and weekly usage limits under Anthropic's Pro subscription tier, is exhibiting a persistent intermittent failure pattern on at least some user systems. The failure manifests as an "Error: Failed to load usage data" message that blocks rendering of the two usage graphs, and critically, once the error appears in a given session window, retrying does not resolve it — the command only begins functioning again after a multi-hour interval. The user reporting this issue is running the latest Claude Code on Node v26 under Debian Linux, suggesting the problem is not obviously tied to an outdated runtime or exotic configuration.
The intermittent, time-bounded nature of the failure strongly implies the issue is upstream of the client — almost certainly a rate-limiting, availability, or caching problem on the Anthropic API endpoint that the `/usage` command queries to retrieve consumption data. When the endpoint is unreachable, throttled, or returning errors, the CLI tool has no fallback mechanism and simply surfaces the raw failure to the user. The fact that it self-resolves after a few hours without any client-side change further supports the theory that the underlying data source goes through periodic degraded states, possibly tied to how Anthropic aggregates or refreshes usage metrics on their backend infrastructure.
This matters beyond mere inconvenience because `/usage` is a core feature for Pro subscribers managing against hard rate limits. Claude Code operates under a usage cap system where developers can be cut off from the tool mid-workflow, making real-time consumption visibility not a nice-to-have but a practical necessity for planning work. When the visibility tool fails precisely during periods of high API activity — which may be exactly when the backend is under stress — users lose the ability to gauge how close they are to interruption, undermining the value proposition of the Pro tier.
The issue also touches a broader pattern in rapidly scaled AI developer tooling where peripheral features like usage dashboards and telemetry are deprioritized relative to core inference capabilities. Anthropic has invested heavily in Claude Code's agentic and coding capabilities, but ancillary infrastructure such as usage tracking endpoints appears to lack the same reliability engineering. The user's pointed remark about the backend being "vibecoded" — a colloquial reference to AI-assisted development that may cut corners — reflects a wider community skepticism about whether the operational maturity of supporting services keeps pace with headline model improvements.
From a technical debt perspective, a robust fix would involve either caching the last-known usage data client-side so the `/usage` command can display stale-but-useful data during endpoint outages, implementing a more graceful degraded mode with informative messaging about the outage, or improving the reliability of the underlying data API itself. The fact that the bug has persisted long enough for the user to characterize it as something they have "been seeing for so long" suggests it has not been prioritized in Claude Code's release cycle, even as the tool has grown in adoption and its user base has expanded to developers with production dependencies on rate-limit awareness.
Read original article →