Show HN: Using Qwen3:1.7B to call itself recursively (seanneilan.com)

A developer has successfully demonstrated the capability of the Qwen3 1.7B language model to execute a multi-step agentic workflow entirely locally. The process involves the LLM recursively calling itself to perform two tasks: first, generating Python code to scrape a webpage and save its content, and second, reading that saved file to summarize its contents. The 1.7B model proved to be significantly more reliable than the smaller 0.6B version, managing to execute tool calls effectively, including generating code and creating "continuation prompts." This experiment highlights the growing sophistication of local LLMs in performing complex tasks autonomously, a significant advancement for the AI/ML community. Leveraging tools defined as Python model classes, the LLM can decide the flow of the task and intelligently choose the necessary code. The success of the Qwen model shows the potential for LLMs to operate as agents capable of executing genuine workflows, laying the groundwork for more advanced applications in AI-driven automation and programming tasks. The developer notes that larger Qwen3 models could further enhance reliability and performance, making them attractive options for future AI development projects.