Skip to content
FSS Knowledge Base

Optimizing AI Benefits

Section titled “Optimizing AI Benefits”

1. Be VERY Clear on Objectives

Section titled “1. Be VERY Clear on Objectives”
  • AI can only help with well-defined goals
  • End important prompts with: “If you need additional information or clarification, ask.”
  • Iterate as needed — first outputs are starting points

2. Maximize Free Daily Usage of Best Models

Section titled “2. Maximize Free Daily Usage of Best Models”

Use free access to best models for ALL learning, research, strategic planning, and tasks independent of long-term dev projects:

Strategy: modest daily use with incubation beats burnout. Assign long research tasks before breaks (meals, bed).

3. Use Deep Thinking/Research Modes

Section titled “3. Use Deep Thinking/Research Modes”
  • For the hardest tasks: research, strategy, planning, dev, debugging
  • Assign long, deep research tasks before breaks from computer

4. Use Best AI Tool for Different Tasks

Section titled “4. Use Best AI Tool for Different Tasks”
  • Use CC in WSL for file/folder/KB work; more efficient
  • Ask Perplexity or Claude: “What AI tools are best for [this task]?”
  • Don’t default to one tool for everything

5. Learn One/Two Best Models Well

Section titled “5. Learn One/Two Best Models Well”
  • Only pay for models that produce results others cannot
  • Most users use a fraction of existing features
  • Claude Code for power users — not just coding (very good at writing)
  • From CC /insights: you’re already a power user (439 sessions/month), but missing key features like hooks, skills, and CLAUDE.md configuration

6. Use Effective Context Engineering

Section titled “6. Use Effective Context Engineering”
  • Instead of vague prompts with weak memory
  • CLAUDE.md, AGENTS.md, .cursorrules — persistent project context
  • Import external docs with @path/to/file syntax
  • Share context between Claude and Cursor via shared config files

7. Use Meta-Prompting

Section titled “7. Use Meta-Prompting”
  • Prompt on how to create a more effective prompt
  • Use built-in prompt improvers from OpenAI, Claude (platform.claude.com)
  • Multi-prompting: send to 3 models, synthesize best insights

8. Ask AI Periodically

Section titled “8. Ask AI Periodically”
  • Ask the best AI models how to optimize AI benefits for your specific needs
  • Run CC /insights monthly and implement recommendations
  • Review and update this document quarterly

9. Evolve a Prompt Library

Section titled “9. Evolve a Prompt Library”
  • For different tasks/roles
  • Store in KB (Obsidian) and in .prompts/ folders in projects
  • See 05_Prompting for frameworks and techniques

10. Implement Verification-First Workflow

Section titled “10. Implement Verification-First Workflow”
  • Your #1 friction: Claude claims “done” without testing
  • Add verification requirements to every CLAUDE.md
  • Build a /taskfix skill encoding the full fix-verify loop
  • Set up hooks for automatic post-edit validation
  • Per-item verification instead of batch execution

11. Build and Use Skills

Section titled “11. Build and Use Skills”
  • Skills are the highest-leverage investment for recurring tasks
  • Use /skill-creator to build skills from patterns you notice
  • Install community skills for common workflows
  • See 02_Skills for the full skills ecosystem

12. Use the Right Model Tier

Section titled “12. Use the Right Model Tier”
  • Haiku for implementation, Sonnet for building, Opus for reviewing
  • GLM-4.7 at $3-6/mn for 80% of daily coding
  • Reserve paid models for what others cannot do