Marketer Zhenya Korzhavin has presented a system of AI agents based on Claude that is capable of independently analyzing advertising campaigns, identifying causes of inefficient spending, and making real-time adjustments to website structures and bid settings.
What Happened
A prototype system has been developed that integrates with the Yandex.Direct and Yandex.Metrica APIs. The agents utilize Claude and MCP (Model Context Protocol) technology to access website source code and analyze user behavior, including block scroll depth. This allows for the automation of the full cycle, from semantic analysis to adjusting landing pages and bids.
Context
The project demonstrates a shift from using LLMs as simple chatbots to the concept of Agentic Workflow. Using Claude's context memory ensures coherence between analysis stages and subsequent changes, allowing the system to function as a single, unified intelligent marketing management loop.
Why It Matters for the Industry
For the industry, this signifies a transition from manual management of fragmented tools to the creation of automated marketing loops. The technology allows for reduced operational time and minimized human error, transforming marketers from interface operators into architects of high-level systems.
Why It Matters for Users
Specialists can build their own stacks of AI agents using ready-made tools like Claude Code and MCP servers without needing to develop base models. This provides the opportunity to radically accelerate reporting preparation and advertising campaign optimization—reducing tasks from several hours to just a few minutes.
What Is Not Yet Known / Limitations
Questions remain regarding the scalability and reliability of the system in production environments. Critical challenges include the complexity of context management as data volumes grow, data security concerns, and the lack of a standardized architecture for such agentic systems.
Sources
Author
Look at AI, Editorial Team