Maximizing Value from Claude Code

The same completed task can cost two very different amounts depending on how you are using Claude Code. In this session, we'll walk through where a session's tokens actually go — what decides the price of a token, what decides how many you send, and the handful of habits and settings that separate an efficient session from a wasteful one.

What you'll learn

• How to pick the right model and effort level. Learn a simple heuristic for each dial: didn't know enough → larger model; didn't try hard enough → higher effort.
• How to cut the cost of long sessions. Learn when to /clear, /compact, and /rewind so you're never paying to re-send a finished task.
• How to get the most out of prompt caching. Identify which changes break the cache (model, effort, fast mode) and the right moments to make them, so your history keeps costing a tenth of input price.
• How to keep your context lean. Quiet noisy commands and hand high-output jobs to subagents so only the answer comes back.
• How to stop paying for startup context you don't use. Learn the ten-minute cleanup that pays off in every session after — run /context to see what CLAUDE.md and MCP servers add to every turn, and cut what doesn't belong.

Featuring

Lydia Hallie, Member of Technical Staff @ Anthropic