Exploring the TTAR
TTAR: Tools Tasking, Actions and Response repsresnets a workwlow for AI. We will break it down below. Tools Tools repsent softare functions and classes which provided an interface and constraints for an AI agent in order for it to accomplish a single task. Tools can consist of access to databases, API calls to Google services etc. These help the again maintain a structure response between reasponing, input and output. Tasking involves promps, prompt engineering, and everything that involves making the initial request for the agent to go out and do something. using the tools, above, the again has gardrails so it doesn't get lost down its rabbit hole. If you think of agents as reasoning machines, you can isolate agents to perform a single aspect of work to reason and work. If a task involves different compontnets. suich as use google maps to find a destination and use a weather service to check on the weather. its best to seperate the tasking among other agents. Actions Actions involve retrtriving, processing and displaying the response using both the tools and tasking above. One major factor for actions includes data formating and presentation. This is important to look at if the action is to be processedd by another tool, or another agent for refinement. JSON format is popular, as its using for both web and non web applications, pretty uiniversially. Response This can consist of either prompt refinments, or as a tools/agent workflow. This may be a decision tree in the middle of a workflow, or as a summary of final results.
Read More