The advisor strategy: Give agents an intelligence boost

2026-04-09 · Source: Claude Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, short

Summary

Anthropic has introduced the "advisor strategy" and a new "advisor tool" on the Claude Platform, enabling developers to achieve near Opus-level intelligence in their AI agents at a significantly reduced cost. This strategy pairs a powerful model like Opus as an advisor with a more cost-effective executor model such as Sonnet or Haiku. The executor handles the primary task, only consulting the advisor for complex decisions. The advisor tool simplifies this by allowing a one-line API change, routing curated context to the advisor model within a single `/v1/messages` request. Evaluations show Sonnet with an Opus advisor increased SWE-bench Multilingual scores by 2.7 percentage points while reducing cost per agentic task by 11.9%. Haiku with an Opus advisor doubled its BrowseComp score to 41.2% and costs 85% less per task than Sonnet solo, offering a strong option for high-volume tasks.

Key takeaway

For NLP Engineers and CTOs building AI agents, adopting the advisor strategy with the new Claude Platform tool can significantly optimize your cost-intelligence trade-off. You should evaluate Sonnet or Haiku with an Opus advisor against your existing benchmarks to confirm performance gains and cost reductions, especially for high-volume or complex tasks where occasional frontier-level reasoning is beneficial.

Key insights

Pairing a powerful advisor model with a cost-effective executor model enhances agent intelligence while significantly reducing operational costs.

Principles

• Escalate complex decisions to a more capable model.
• Isolate high-cost reasoning to critical junctures.

Method

Configure a smaller executor model (Sonnet/Haiku) to perform tasks and invoke a larger advisor model (Opus) only when it encounters decisions it cannot reasonably solve, within a single API request.

In practice

• Use `advisor_20260301` in Messages API for cost-effective agents.
• Set `max_uses` to control advisor calls and manage spend.
• Integrate with existing tools like web search or code execution.

Topics

• Advisor Strategy
• Claude Platform
• AI Agent Architectures
• Cost Optimization
• Large Language Models

Best for: NLP Engineer, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, AI Architect

Related on AIssential

• Harnessing Claude’s intelligence — Building with evolving AI requires adapting agent harnesses to leverage new model capabilities and optimize performance.
• not much happened today — AI agentic workflows are maturing with advisor-executor patterns, robust harnesses, and more realistic evaluation benchmarks.
• Anthropic's Code With Claude Announces Managed Agents, Proactive Workflows, Capability Curve — Model step-changes fundamentally reshape product architecture, organizational design, and infrastructure economics.
• The Business Engineer's Claude OS — Integrating structured mental models and analytical engines enhances AI for strategic business analysis.
• Introducing Claude Opus 4.8 — Claude Opus 4.8 significantly enhances agentic capabilities, honesty, and alignment, offering improved performance and new features at existing or reduced costs.
• Claude Skills and Subagents: Escaping the Prompt Engineering Hamster Wheel — Claude Skills and subagents offer a lazy-loading, modular approach to AI agent development, optimizing context management and token economics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Claude Blog.