Advisor Mode: How to Use Opus as a Strategist and Sonnet as the Builder

If you use Claude Opus for everything in Claude Code, you are burning through your token budget at the worst possible rate. Here is the reality: Opus costs roughly 5 times more per token than Sonnet. And in a typical coding session, up to 80% of tokens are spent on execution — writing functions, generating files, applying diffs — tasks that Sonnet handles just as well.

Advisor Mode (also called Opus Plan Mode) solves this by splitting the cognitive labor. You activate it with a single command: /model opus-plan. From that point, Opus handles the thinking and architecture — understanding your codebase, tracing dependencies, designing the plan. Then Sonnet takes over for execution, writing the actual code line by line at a fraction of the cost.

Alternatively, you can use the /advisor toggle to keep Opus available as an on-demand reviewer while Sonnet keeps executing — same savings, more flexibility.

The math is brutal in the best way: a session that used to hit the limit after generating a few hundred lines of code can now handle thousands of lines before reaching the same wall. The planning phase with Opus costs just a few hundred tokens. The savings on execution are massive.

Follow for more AI productivity tips:

  • https://www.instagram.com/ai.with.mo/

    • Course Registration: https://tally.so/r/D4KBB5

    The Real Cost of Using Opus for Everything

    Claude Opus costs roughly 5 times more per token than Sonnet at the API level. In a typical Claude Code session, up to 80% of token consumption happens during execution — writing functions, generating files, applying diffs. This is mechanical work that does not require Opus-level reasoning. When you leave Opus running for the entire session, you are spending premium budget on tasks a cheaper model handles just as well. A session that burns through its limit after generating a few hundred lines of code could handle thousands of lines if you routed execution through Sonnet instead. The planning overhead is minimal — a few hundred tokens at most. The savings on execution are massive.

    How Advisor Mode Works: Split the Cognitive Labor

    Advisor Mode splits your workflow across two models based on what each does best. Activate it by typing /model opus-plan in Claude Code. From that point, Opus handles the high-value thinking: understanding your codebase structure, tracing dependencies across files, identifying edge cases, and producing a detailed execution plan. Then Sonnet takes over for the actual implementation — writing code, creating files, applying diffs — at a fraction of the cost. Switching models does not clear the conversation, so Sonnet sees everything Opus reasoned through. Alternatively, use the /advisor toggle to keep Opus available as an on-demand reviewer while Sonnet continues executing. Same savings, more flexibility for ongoing sessions.

    The Right Way to Use It: Review the Plan Before You Execute

    The biggest mistake people make with Advisor Mode is treating the plan as a formality and immediately saying 'go ahead.' Plans can have real errors. Opus might misunderstand your codebase structure, plan changes to the wrong files, or over-engineer a simple problem. Before approving any plan, read it carefully. Correct misunderstandings in plain English — 'skip the legacy folder,' 'do not touch the existing tests,' 'start with the middleware layer first.' Also, describe your full task in one detailed prompt rather than one change at a time. Opus will handle related changes as a coherent plan, and Sonnet will execute them in sequence. This is far more efficient than repeated back-and-forth, and it dramatically reduces the total token cost of the session.

    Advanced Tips: Maximize Your Token Budget

    Beyond Advisor Mode, several habits compound your savings significantly. Use /effort low when asking Opus to plan straightforward tasks — same model, far less overthinking and token burn. Use /compact between phases to trim the conversation history and free up context window space. Run /cost regularly to monitor your spend before starting large tasks. Reference file paths instead of pasting file contents — 'look at the validateToken function in src/auth.ts' costs a fraction of pasting the entire file. For logs and stack traces, trim to the 20-30 relevant lines before sharing. And if a task is purely mechanical — renaming a variable, adding a comment, writing a quick regex — skip both Opus and Advisor Mode entirely and use Haiku directly. Match the model to the task, not the other way around.

    Prompt

    # ADVISOR MODE — TWO WAYS TO ACTIVATE

# Option 1: Opus Plan Mode (recommended for large builds)
# Opus plans the full roadmap, Sonnet executes every file
/model opus-plan

# Option 2: Advisor Toggle (recommended for ongoing sessions)
# Keep Sonnet executing, call Opus as an on-demand reviewer
/advisor

# ─────────────────────────────────────────
# RECOMMENDED WORKFLOW:

# Step 1 — Describe the full task to Opus in one detailed prompt
# (batch related changes together — do NOT do one file at a time)
"Build a full authentication system with JWT, refresh tokens,
and role-based access control. Divide into phases and plan
which files you will touch in each phase before writing any code."

# Step 2 — Review the plan carefully before saying go ahead
# Correct misunderstandings in plain English:
"Skip the legacy/ folder. Do not touch existing tests.
Start with the middleware layer first."

# Step 3 — Let Sonnet execute the approved plan
# Sonnet sees everything Opus produced — no context lost

# ─────────────────────────────────────────
# PRO TIPS:
# - Use /effort low for Opus on straightforward planning tasks
# - Use /compact between phases to trim conversation history
# - Use /cost to monitor token spend in real time
# - Never paste entire files — reference paths instead (saves tokens)
# - Haiku is available for simple lookups and quick edits