The video introduces Claude’s new advisor strategy that dynamically selects between powerful and smaller AI models like Opus, Sonnet, and Haiku to optimize performance and reduce costs, though using Opus with an advisor can be significantly more expensive. The creator demonstrates testing tools, setup instructions, and shares personal preferences while promoting their broader AI ecosystem and community resources.
In this video, the creator introduces Claude’s newly released advisor strategy, designed to make AI agents smarter while reducing costs. The advisor strategy allows Claude to decide when to use a more powerful model like Opus versus smaller models such as Sonnet or Haiku, aiming to optimize performance and cost-efficiency. Opus is the most expensive and powerful model, roughly twice the cost of Sonnet, which is mid-range, while Haiku is the smallest and cheapest. The advisor strategy is intended to balance usage between these models to improve overall efficiency.
The creator demonstrates a tool they developed to test and compare costs and performance across different model configurations, including solo models and those combined with the Opus advisor. Their tests reveal that using Opus with an Opus advisor can be significantly more expensive—sometimes up to eight times the cost of using Opus alone—especially for simple tasks like greetings. Despite the higher cost, the advisor strategy can yield better performance, with some configurations showing up to 2.7% improvement and better results on benchmarks like software engineering tasks.
Latency is another factor explored in the video, with the advisor strategy sometimes increasing response times, though not consistently across all tests. The creator runs live tests with example prompts, such as explaining the CAP theorem and designing a distributed cache, to compare outputs and costs. They note that Haiku combined with an Opus advisor might offer a good balance of cost and performance for less complex tasks, while Opus alone or with an advisor is more suited for demanding applications.
The video also covers how to set up the advisor strategy within Claude Code, using commands like /model to select a base model and /advisor to configure the advisor model. The advisor helps delegate tasks between models, potentially improving accuracy and efficiency. However, the creator personally prefers to use the most powerful model directly without an advisor, as they tend to maximize their usage limits and prioritize performance over cost savings.
Finally, the creator mentions their broader ecosystem, including a platform called Arc for managing websites, applications, and AI agents in one place. They promote their paid community, which offers resources like a five-day Claude challenge for beginners, masterclasses, and downloadable applications that integrate seamlessly with Arc. The video concludes with an invitation for viewers to share their thoughts on using the advisor mode and encourages engagement through likes, comments, and subscriptions.