The simple case for AI agents
So, considering the use of the AI Agents, there is concept regarding harness engineering and it is truly important.
The discussion usually is regarding the necessary layers that are required for great work for the model. But, also, it is important to consider unnecessary layers or strategies so that the model can be not atrapalhado. Like, let the model cerceado por um caminho and ask the model to reason for steps, while the model could just know how to reason fast and find the result, it can not be the best strategy.
Like, trust the model, so, observe the results and provide what is necessary. The harness engineering is also about building based on what the model needs. Interesting.
Really great learning.