QuestionQ22

VerifiedDeveloping custom agents

You are developing an autonomous financial research agent using Agent Development Kit (ADK). The agent has access to a registry containing more than 50 distinct financial analysis tools. When you bind all 50 tools to the agent at once, the system prompt consumes too many tokens, causing high latency, increased costs, and frequent LLM hallucinations in which the agent selects the wrong tool schema. You need to improve the agent’s tool-calling accuracy and reduce token consumption while ensuring all 50 tools remain available for the workflow.

What should you do?

Choose two
Explanation

Semantic retrieval can rank tool descriptions against the current prompt and select a relevant top-K subset. Binding only that subset reduces the number of schemas sent to the model and narrows its selection, while the full registry remains available for subsequent retrieval. ADK supports filtering exposed tools; semantic retrieval can be implemented as a custom component rather than assumed to be a built-in ADK feature. The subset may need to be refreshed if a workflow requires additional tools. (adk.dev)

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!