AI — Claude Code Tips

AI — Claude Code Tips

10 Tips from the Claude Code Team

1. Parallel Sessions (Git Worktrees)

Use git worktree to run multiple Claude sessions on the same repo simultaneously:

  • Main repo → Worktree 2a (Claude Session 1), 2b (Claude Session 2), 2e (Claude Session 5)
  • Each session works in isolation without blocking others
  • Dramatically increases throughput on large tasks

2. Plan Mode for Complex Tasks

  • Claude 1 writes the plan → Claude 2 reviews (Staff Eng perspective)
  • If sideways → re-plan
  • Invest in planning before coding. Also useful for verification.

3. Treat CLAUDE.md Like a Memory System

  • After every correction, tell Claude: “Update CLAUDE.md so you don’t make that mistake again.”
  • Edit ruthlessly over time — mistake rate actually drops
  • CLAUDE.md is the persistent memory system across sessions

4. Turn Repeat Work into Skills

  • Any task done >1x/day → turn into a Custom Skill (Command/Agent)
  • Examples: /techdebt to kill duplicated code, Sync Slack/GDrive/Asana/GitHub in one context, Analytics agents (dbt, reviews, tests)
  • Automate frequent tasks

5. Autonomous Bug Fixing

  1. Paste Slack bug thread
  2. Say “fix”. Point at logs.
  3. Use Claude autonomously.
    • Inputs: Slack bug thread + Failing CT tests + Docker logs

6. Harsh Reviewer

  • Share your changes (diff)
  • Ask Claude to “Grill me” / “Prove this works” / “Scrap this and implement elegant version”
  • Demand rigorous reviews before PR

7. Terminal Setup Matters

  • Ghostty terminal
  • /statusline (Contact Usage + Branch)
  • Voice Dictation (~3x faster)
  • Optimise environment with tools

8. Use Subagents

  • Subagent 1 (Narrow Task) + Subagent 2 (Narrow Task) → Main Claude
  • Offload tasks to keep context
  • Route permissions to Opus via hooks

9. Claude for Analytics

  • Claude (SQL Skill) → BigQuery CLI, Any DB with CLI/MCP/API
  • One engineer hasn’t written SQL in 6+ months using this pattern

10. Learn with Claude

  • Explanatory output (/config)
  • HTML Presentations
  • ASCII Diagrams
  • Spaced-Repetition Skill
  • Explain understanding; Claude fills gaps, stores result; master new concepts

Cursor vs VS Code — Decision Guide

Source: MyLearn/AI/rbk-queries-during-ai-discovery-phase.md

Why Choose Cursor over VS Code?

Cursor is built on the VS Code codebase — it retains a familiar interface while optimising for AI-first coding. Choose based on your primary workflow:

Choose Cursor when:

  • You want deep, native AI integration (not extensions bolted on)
  • You need project-wide understanding — AI analyzes the entire codebase for multi-file edits
  • You want dedicated chat, inline diffs, and “Composer” mode (generate projects from prompts)
  • You want multi-model support (switch between OpenAI, Anthropic etc. in one session)
  • You want AI-assisted collaboration: shared chats, rules, BugBot code reviews
  • You want stronger tab completion (more predictive multi-line suggestions)
  • You’re a beginner who wants code explained like a coaching session

Choose VS Code when:

  • You value maximum flexibility through extensions
  • You work across many languages/ecosystems with specific extension needs
  • You prefer a non-AI-centric workflow with AI as an optional add-on
  • Your team has standardised on VS Code tooling

In short: VS Code = flexibility + ecosystem. Cursor = streamlined AI-centric workflow.

See Also

Trending Tags

blogging