Methods

Methods are specialized workflows or information processing techniques that LLMs follow. These Methods provide a means to externalize mental processes, utilizing various Node types in the Coordination Editor to perform actions in an automated manner. They process Nodes in a stepwise fashion, with the user maintaining control over the process. Unlike Retrieval Augmented Generation (RAG) workflows, Methods offer granular control over data processing while ensuring full auditability. The Coordination Editor will feature a marketplace where these Methods can be shared and utilized by others.

A key aspect of Methods is their intuitive visual design, which allows experts to adapt them in real-time as they work. This flexibility enables continuous refinement, making the Methods increasingly effective over time. As experts modify and improve these Methods, others can benefit from their accumulated wisdom. Methods are not standalone entities but serve as building blocks that can be integrated with other Methods.

The creation and integration of these Methods fall under the purview of a new role: the "AI Business Engineer". This role involves codifying an expert's thought processes into a Method, bridging the gap between traditional technical engineers and domain experts. Once a Method is constructed and tested, it should be intuitive enough for use without requiring constant involvement from the AI Business Engineer. Importantly, experts can continue to refine these Methods, leading to continual optimization.

With the release of the GPT-o1 Model, the Coordination Editor will add functionality to set processing of Nodes with different models. For example, users can specify the use of GPT-o1 for critical thinking tasks and Claude 3.5 Sonnet for writing tasks within the same Method. This allows for optimization of different steps in the process based on the strengths of each model.

Looking to the future, it's anticipated that more advanced models will be able to automatically select which LLM is the optimal choice for each specific task within a Method. This automatic selection will further streamline the process, reducing the need for manual configuration and allowing for more efficient and effective workflows.

The Methods described in this document provide an initial framework to achieve the desired outcome. We envision significant changes to these Methods over time. As models improve, the need for detailed step-by-step outlines may decrease leading Methods to become simpler in structure and more integrated.