The Agentic Loop: Three loops in a trench coat
bobbytables.io
July 14, 2026
6 min read
🔥🔥🔥🔥🔥
46/100
Summary
Agent loops consist of three distinct components rather than a single loop, creating a more complex "agentic" experience for users. The discussion includes examples of pseudo-code to illustrate these concepts.
Key Takeaways
- The agentic experience for customers is composed of three distinct loops: the inference loop, the tool loop, and an unspecified third loop.
- The inference loop is responsible for making chat completion API calls, managing chat history, and passing tool usage requests.
- Large Language Models (LLMs) operate on a stateless design, requiring the entire conversation context to be sent with each API request.
- The tool loop enables LLMs to utilize external tools, enhancing their functionality beyond mere text generation.
Community Sentiment
Positives
- The pattern of three loops is a compelling framework that could help clarify the interaction between human input and AI processes.
- Commenters are actively engaging with the concept, indicating a thriving interest in refining AI models and their architectures.
Concerns
- There's skepticism about the effectiveness of these loops for complex tasks, with one commenter noting that agents often reward hacks instead of delivering generalized solutions.
- Concerns about the loops being misrepresented in the diagram highlight a lack of consensus on the model's structure and functionality.
