Context Budget

A context budget is the deliberately planned set of information given to an AI model or coding agent for a specific task. It includes the system prompt, project rules, relevant files, examples, tickets, error logs, tool outputs, and the history of previous steps. Because every model has a limited context window, the budget determines whether the agent can reason from the right evidence or gets distracted by noise. Strong teams treat the context budget as an engineering artifact: they select sources, rank hard requirements above background material, remove irrelevant files, and keep enough traceability for review. In agentic workflows, context budgeting is also a cost and safety control. Smaller, better curated context lowers token spend, reduces accidental data exposure, and makes results easier to reproduce. Too little context, however, creates hallucinations, wrong assumptions, and avoidable back-and-forth. In practice, context budgeting means clarifying the task, packaging only the needed evidence, documenting intermediate results, and refreshing context deliberately during long-running agent work.