Allow ParallelAgent to selectively run only relevant subagents based on query

[jhammarstedt]

The current implementation of ParallelAgent executes all its sub-agents concurrently, regardless of the specific requirements of the input query. This design is optimal for scenarios where tasks are entirely independent and all sub-agents are relevant. However, in many real-world applications, only a subset of sub-agents may be pertinent to a given input, leading to unnecessary resource utilization and potential inefficiencies.

Proposal:

Introduce a mechanism within ParallelAgent to dynamically determine and execute only the sub-agents relevant to the current input context or let the parent of the ParallelAgent be able refer to the sub agents which would trigger only these to be called

Is there an existing pattern or recommended approach within ADK to achieve this dynamic sub-agent selection? If not, could it be considered as a feature? Looking at other providers like Antropic they [support and encourages this behaviour] for their multi agent research workflow (https://www.anthropic.com/engineering/built-multi-agent-research-system)