Using AI Agents Efficiently

A collection of practical tips for getting more out of AI coding assistants — without burning through tokens unnecessarily.

Images

Pasting screenshots is expensive. Each image consumes significant tokens for vision analysis.

• Describe layouts in text instead — “a 3-column grid with a hero section above” costs far less than a screenshot
• Use images only for details that are genuinely hard to describe (custom icons, specific gradients, unusual UI patterns)
• Reference existing code rather than screenshots of your own UI — paste the component path or code snippet directly

Be Specific and Targeted

• Ask about one file or function at a time rather than broad questions like “fix my whole codebase”
• Reference specific line numbers: “in page.mdx around line 40, how do I…”
• Include the “why” so the AI doesn’t need to ask: “I need X because Y, so don’t suggest Z”

Avoid Redundant Context

• Don’t paste large files when you only need help with a small section — select just the relevant portion
• In JetBrains AI, open files are often auto-included as context — close files you’re not actively working on

Use the Right Tool

• For simple completions (autocomplete, boilerplate), use inline completion rather than chat
• Chat sessions are heavier — reserve them for reasoning tasks, not quick lookups
• File search, renaming, extracting, and formatting don’t need AI — use the IDE’s built-in tools for those

Manage Chat Sessions Wisely

• Continuing an existing chat is more efficient than starting fresh when the context is still relevant
• Start a new session when switching topics — long chats filled with resolved questions add noise and cost

Write Clear Prompts the First Time

• Complete, well-formed prompts avoid the costly back-and-forth of clarification
• State your constraints upfront so the AI can give a useful answer on the first try