Quick answer
The local whoburnedmore CLI command is npx whoburnedmore --local. Decline its optional publish offer to keep the report local. The standard command signs in and submits daily aggregates. Neither mode submits prompts, source code, or filenames.
What this guide answers
The reader wants a practical template for managing AI coding spend.
Best for
Individual developers, team leads, and founders managing usage-based AI tools.
Key facts, checked November 8, 2026
- Anthropic's figure for Claude Code on enterprise deployments is about $13 per developer per active day and $150-250 per developer per month, with 90% of users below $30 per active day. That is a reasonable first cap for an agent-heavy developer. source
- GitHub Copilot includes 1,900 AI credits ($19) per Business user and 3,900 ($39) per Enterprise user each month, pooled across the organization. source
- DORA (June 2, 2026) says company-wide token aggregates are meaningless and individual scores are harmful. It recommends tracking at the team, application or use-case level, alongside cost per accepted change. source
- Cursor's spend limits are not enforced instantly, so usage can briefly exceed a limit. Leave headroom between your warning line and the hard cap. source
Template fields
Keep the template simple. Each row should describe a tool, budget window, threshold, owner, and the task classes it is meant for.
- Tool
- Weekly cap
- Warning threshold
- Owner
- Allowed tasks
Use task categories
Task categories prevent moral arguments about usage. A high-burn production fix is different from a high-burn styling loop.
- Incident
- Feature
- Review
- Experiment
- Cleanup
Add exception rules
A budget without exceptions will be ignored during urgent work. Write the exception path before the urgent work happens.
- Who approves
- What to record
- When to review
Review the top burns
The weekly review should produce one behavior change. Do not spend an hour debating every prompt.
- Largest useful burn.
- Largest wasted burn.
- One rule to change.
A filled-in example row
Copy this layout and fill it in per tool. Tool: Claude Code on Max ($100/month). Window: weekly, matching the plan's weekly limit. Expected use: about 60% of the weekly limit. Warning: 80% on /usage. Owner: the developer. Allowed tasks: feature work, refactors, incident fixes. Exception: an incident can exceed the limit, with a one-line note in the review doc. Outcome column: merged PRs for the week. For an API-billed tool, swap the plan percentage for dollars, using Anthropic's $13 per active day enterprise average as a starting point.
- Tool and plan
- Window that matches how the plan resets
- Expected use and warning line
- Owner and allowed tasks
- Exception rule
- Outcome measure
Match each window to the vendor's reset
A budget that resets monthly while the tool resets weekly will always look wrong. Claude Pro and Max use a 5-hour session limit that resets in full five hours after your first message plus a weekly limit, and Ultra has only the weekly limit. Copilot AI Credits reset on the 1st of the month (UTC). Cursor's included usage is a weekly pool. Lovable gives 5 daily credits on top of monthly credits. Set each row's window to match its tool, then roll everything up into one monthly dollar figure for finance.
Fill the actuals column automatically
For local coding agents, npx whoburnedmore reads token logs from 22 supported tools and shows daily totals by tool and model with estimated cost. With install-sync it refreshes every 15 minutes. Paste the week's top three burn days into the review. For cloud tools it can't read, such as Replit, Lovable and Devin, copy the numbers from each vendor's usage page.
Common questions
How much should I budget for AI coding per developer?
Anthropic reports about $150-250 per developer per month for Claude Code in enterprise deployments. Seat plans such as Copilot Business at $19 or Claude Pro at $20 set the floor, and agent-heavy use costs more.
Should AI token budgets be per person or per team?
DORA recommends measuring at the team, application or use-case level, because individual token scores invite gaming. Keep personal caps as guardrails, not scorecards.
What period should an AI budget cover?
Match each tool's reset window, such as weekly for Claude limits and monthly for Copilot credits. Then report a combined monthly total.