Student developer shares three habits to cut AI coding agent costs

A recent undergraduate, now applying to Master's programs, has outlined practical strategies to reduce token usage when working with AI coding agents like Claude Code and Kiro. The author argues that every prompt, model response, and tool call consumes tokens that cost real money, making cost control especially important for students on tight budgets. A core recommendation is to arrive at each session with a clear plan, specifying known requirements upfront so the agent does not waste tokens rediscovering decisions already made. The writer also advises distinguishing between parts of a task that need straightforward execution and parts that require the agent to reason, since reasoning draws more tokens. Choosing frugal models for simpler tasks and crafting well-scoped prompts, the author suggests, can reduce token consumption to roughly a tenth of what a vague, open-ended request would require.