Mastering Claude Code Token Efficiency

Claude Code efficiency is driven by managing the context window and maximizing prompt caching to lower token costs. Users should initialize session settings early, use @-mentions for files, and minimize command output to keep the conversation history focused. Regularly clearing or compacting sessions prevents historical data from inflating the cost of future turns.

Key Points

• Prompt caching is the most effective way to reduce costs, but it is broken by mid-session changes to models, effort levels, or long periods of inactivity.
• Output tokens are significantly more expensive than input tokens because they require sequential GPU processing during the decode phase.
• Every file read and command output stays in the session context, meaning users pay to re-send that data on every subsequent turn.
• Strategic use of commands like /clear, /compact, and /rewind helps maintain a relevant context window and prevents the model from becoming distracted by irrelevant data.
• Subagents can be used to isolate high-output tasks, returning only the necessary information to the main session to save on context space.

Sentiment

The community mostly accepts the article's core premise that token and context discipline are important for agentic coding, and many commenters contribute practical refinements from their own workflows. At the same time, the reaction is noticeably skeptical of the product burden placed on users: several commenters argue that cache-aware compaction, context separation, and session management should be first-class tool behavior rather than manual rituals. Hacker News therefore leans supportive of the techniques but critical of the current ergonomics and incentives around them.

In Agreement

• Frequent fresh sessions with explicit handoff files can preserve the important project state while avoiding long, noisy contexts.
• Portable markdown plans, ledgers, and docsets are valuable because agents and humans can inspect, reuse, and transfer them across tools.
• Keeping sessions short, pruning loaded instructions, using quiet tool output, and reviewing where tokens are wasted are practical ways to improve agent performance.
• Prompt caching and model or effort selection are real operational concerns that users need to understand when cost and session quality matter.
• Subagents and task-specific plans can reduce context bloat by isolating noisy work from the main conversation.

Opposed

• Claude Code should handle more of this automatically instead of requiring users to manually compact, clear, hand off, and manage cache timing.
• Clearing context may not always be cheaper or better when the same codebase context will simply need to be reread and re-taught.
• Some recommendations depend on product features that commenters find unreliable, such as file @-mentions in certain Claude interfaces.
• Token-efficiency guidance can feel like usage austerity or vendor cost management rather than user-centered product design.
• Compression plugins and terse prompting schemes may introduce enough indirection and failure that the saved tokens are not worth the tradeoff.