Module 4: Token Optimization & Cost Management - DRAFT

Why Token Management Matters

• Tokens drive cost.
• Large models are significantly more expensive than lighter ones.
• Poor habits (long chats, verbose prompts, unnecessary context) quietly inflate spend.

---

Right Tool, Right Task

Examples (adapt for your stack):

• Research & fact-finding → Perplexity (public info, cheaper).
• Deep analysis & narrative → Claude.
• Simple extraction/classification → smaller/cheaper models where available.

Avoid defaulting to the highest-cost model for every task.

---

Prompt & Conversation Practices

Prompts

• Use short, clear instructions.
• Specify structure (bullets, table, JSON) to keep outputs lean.
• Remove conversational filler.

Conversations

• Start a new thread when the topic or phase changes significantly.
• Summarize prior context into a concise, reusable prompt.
• Avoid dragging a huge history into every call.

---

Skills, Connectors & Context

• Only enable skills/connectors needed for a task.
• Turn off integrations that are not in use.
• For RAG or document queries, avoid loading entire corpora; retrieve only relevant slices.

---

Employee Checklist (Cost)

• I can explain why I picked a given model for a task.
• I keep prompts concise and structured.
• I start fresh threads with summaries instead of dragging old context forever.
• I minimize unnecessary skills/connectors.