Protect the important days

If you tend to burn most tokens early in the week, important late-week work can suffer. A weekly view helps you reserve capacity for launches, reviews, and incidents.

• Mark release days.
• Avoid broad experiments before deadline work.
• Keep a fallback tool for low-value prompts.

Classify work by leverage

Not every AI task deserves the same budget. Deep implementation, unfamiliar code, and debugging may justify high burn; simple search and formatting usually do not.

• High leverage: production fixes.
• Medium leverage: refactors with tests.
• Low leverage: style churn and repeated rewrites.

Review burn patterns weekly

The point of measurement is behavior change. A weekly review should identify one waste pattern and one high-value use case worth repeating.

• Find your largest session.
• Ask whether it shipped value.
• Turn the lesson into a team rule.

Use public accountability carefully

A leaderboard can make burn visible, but it should not reward waste. Pair token totals with outcomes so the team celebrates useful leverage, not empty consumption.

• Show shipped work beside burn.
• Avoid ranking people without context.
• Use team boards for friendly comparison.

Find your reset time and pace

Your weekly reset is fixed per account, so look it up once and plan around it. Settings > Usage on claude.ai shows the next reset time. In Claude Code, a status line script can read rate_limits.seven_day.used_percentage and rate_limits.seven_day.resets_at, which is enough to show a pace meter. For a quick check, divide the percentage used by the share of the week that has passed. A result above 1.0 means you will run out early.

• Remember that claude.ai chat draws from the same weekly pool as Claude Code.
• The /usage breakdown has a 7-day view (press w) showing which skills, subagents, MCP servers or loops used the most.

When the weekly limit hits before your work is done

"You've hit your weekly limit" applies to all models, so switching with /model will not help. If your plan has a Fable-specific weekly limit and you hit that one, switching to Opus or Sonnet keeps you working. Usage credits let work continue past the limit, and on Pro, Max and Ultra you can set a monthly spend limit for them under Settings > Usage. The monthly API credits that come with Max and Ultra cannot be used in Claude Code or the Claude apps.

Common weekly drains to cut first

Anthropic's cost guide lists the patterns behind runaway weekly usage. Long sessions resend their full context on every turn. /loop and scheduled tasks fire even while you are idle. Agent teams use roughly 7x the tokens in plan mode. Opus left as the default model, together with subagents that inherit it, multiplies cost. Fix those before deciding you need a bigger plan.