10 Productivity Tips from the Claude Code Team

Boris Cherny, one of the creators of Claude Code, recently shared his team's top 10 tips for getting the most out of the tool. These come directly from the people who build and use Claude Code every day, and several of them go well beyond what most users are doing.

Run Multiple Sessions in Parallel

The single biggest productivity unlock, according to the team, is running three to five Git work trees with each one running its own Claude Code session. Instead of waiting for one task to finish before starting the next, you spin up parallel sessions and let them all work simultaneously.

Start Complex Tasks in Plan Mode

Before jumping into implementation, switch to plan mode and invest your energy in getting the plan right. Once Claude has a solid plan, it can often one-shot the implementation. Some team members even spin up a second Claude session to review the plan like a staff engineer before executing it.

Invest in Your Claude.md File

Every time you correct Claude, end the conversation with "update your Claude.md so you don't make that mistake again." Claude is good at writing rules for itself. Over time, this file becomes a living document of project-specific knowledge and preferences that prevents repeated mistakes.

Create Custom Skills and Commit Them to Git

If you do something more than once a day, turn it into a skill or slash command. One team member built a tech debt command that runs at the end of every session to find and eliminate duplicated code. These custom commands live in your repo and can be shared across the team.

Let Claude Fix Bugs Autonomously

Enable the Slack MCP, paste a bug thread, and just say "fix." Or point it at failing CI tests and let it work. Stop micromanaging the path it takes to get there.

Level Up Your Prompting

Challenge Claude. Say things like "grill me on these changes, and don't make a PR until I pass your test." Use it as a reviewer. Or say "prove to me this works" and have it diff behavior between branches. The more you push back on Claude, the better your results.

Use Sub-agents for Parallel Compute

Append "use sub-agents" to any request where you want Claude to throw more compute at the problem. It keeps your main context window clean while offloading tasks to run in parallel. The frame below shows five sub-agents exploring different parts of a codebase simultaneously.

Terminal Setup, Data Work, and Learning

The team recommends Ghostty for its rendering and Unicode support, and suggests using the status line command to always show context usage and current Git branch. They also use Claude for data and analytics directly, with a BigQuery skill checked into the codebase. The creator says he personally hasn't written a line of SQL. Finally, enable the explanatory output style in config so Claude explains the reasoning behind its changes, turning every session into a learning opportunity.

Key Takeaways

• Parallel sessions across Git work trees is the team's top productivity recommendation
• Your Claude.md file should be a living document that improves with every correction
• Sub-agents let you throw more compute at problems without polluting your main context
• Custom skills and slash commands eliminate repetitive workflows
• Use Claude as a reviewer and challenger, not just an implementer

Resources

• Boris Cherny on X -- Claude Code creator who shared the original tips