Tokens, Context, and Why Small AI Tasks Aren't Cheap

Short summary

Using Cursor Agent Mode for a simple font-pairing task cost 1% of monthly API budget because agents re-read all tagged files across multiple loop iterations (read, plan, edit, verify). Input tokens are cheaper than output but accumulate fast; even a focused agent loop typically makes 4-5 API calls, each passing your full context. Key takeaway: decide if you need the AI to act or just provide an answer—for simple tasks, Ask mode costs a fraction of the tokens.

• •Agent Mode re-reads context across multiple loop iterations, making small tasks surprisingly expensive
• •A single agent loop typically involves 4-5 API calls, each passing the full tagged file context
• •For simple questions or single-file edits, Ask Mode costs a fraction of the tokens

Generated with AI, which can make mistakes.