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 to understand why large context windows increase AI coding cost.
Best for
Developers using Claude Code, Codex, Cursor, Gemini CLI, and similar tools.
Context is the working set
The model can reason only over what it sees, but every extra token competes for budget. Give enough context to solve the problem, not the whole repository.
- Relevant files
- Failing output
- Constraints
- Acceptance test
Long sessions accumulate weight
A session that starts focused can become expensive as history grows. Summaries and fresh sessions help when the task changes.
- Reset after task shifts.
- Summarize decisions.
- Drop obsolete files.
Logs are expensive
Pasting full logs is often unnecessary. Give the failing command, important error, and a short excerpt before sending thousands of lines.
- Command
- Error
- Relevant stack
- What changed
Measure before optimizing
If a workflow is costly, compare the same task with narrow context and broad context. The difference teaches better prompt defaults.
- Broad run
- Narrow run
- Outcome comparison
Mistakes to avoid
Optimizing before measuring
It is tempting to change plans, switch tools, or clamp down on usage as soon as model context window cost becomes a concern. That usually hides the real issue. Measure the current workflow first, then decide whether the problem is volume, scope, model choice, team policy, or one unusually expensive session.
Comparing vendor units directly
A request, credit, ACU, message, token, and quota are not interchangeable units. Convert each tool back to the work it produced: the feature, bug fix, review, prototype, or incident response. That makes cross-tool comparison fair enough to act on.
Treating high burn as automatically bad
A high-burn session can be waste, but it can also be the session that unblocked a release. Add outcome notes before judging the number. The goal is not low usage; the goal is useful, explainable usage that the team can repeat.
Practical playbook
What to measure first
Start with the signal most likely to change behavior for this topic: files included. For someone searching model context window cost, the useful answer is not a generic definition. It is a repeatable way to decide whether the current workflow is healthy, whether the cost is justified, and which next action will reduce waste without killing useful AI experimentation.
How to turn it into a habit
Use a simple weekly rhythm: measure the biggest burn, label the task, record whether it shipped value, and change one prompt or routing rule. The sections above cover context is the working set, long sessions accumulate weight, logs are expensive, and measure before optimizing. Those are the pieces that make the guide actionable instead of another pricing summary.
How whoburnedmore fits
whoburnedmore is the measurement layer, not the policy layer. It reads local AI coding-agent usage, keeps source code out of the upload path, and gives you a shared burn view. That means this guide can stay focused on decisions: when to upgrade, when to narrow context, when to switch tools, and when a high-burn session was actually worth it.
Decision checklist
Can you explain why model context window cost matters for a real task this week?
Do you know which tool, model, project, or workflow created the largest burn?
Is the next action a smaller prompt, a different tool, a plan change, or a team policy update?
Can you review the result without uploading source code or raw prompt content?