An insurance company built two separate Foundry agents: an intake agent that gathers claim details from customers, and a fraud-review agent that scores claims for risk. The team wants the intake agent to automatically call the fraud-review agent mid-conversation to get a risk score before finalizing the claim, without a developer manually wiring a custom integration between the two chat sessions.
Select an answer to reveal the explanation.
Short Explanation
When you've already got two agents built for two different jobs and you want one to just reach out and use the other mid-conversation, the platform-native way is to connect them so one agent can call the other directly, no custom glue code required. Copying transcripts back and forth by hand defeats the whole point of wanting this automated, that's manual labor standing in for an integration. Smashing both agents into one giant system prompt throws away the specialization you built each one for, you lose the clean separation between gathering claim details and scoring fraud risk. And having one agent poll the other's database on a timer means the risk score shows up late, after some delay, instead of right there in the live conversation when it's actually needed. Direct agent-to-agent calling is what gets you the immediate, automatic handoff.
Full Explanation
The correct answer is D. Foundry Agent Service supports connecting one agent to another as a connected agent, letting the intake agent call the fraud-review agent directly within the same conversation to get a risk score in real time, exactly the automatic, mid-conversation call the team wants without custom integration work. Option A is incorrect because manually copying transcripts between sessions is a manual, error-prone process that defeats the goal of automation and does not scale to a production claims volume. Option B is incorrect because merging both agents into one system prompt discards the specialized behavior and separation of concerns each agent was built for, and it is not how the team wants to preserve two distinct, purpose-built agents. Option C is incorrect because polling a database on a fixed schedule introduces delay between when a claim is submitted and when it gets scored, rather than getting an immediate risk score during the live conversation as the scenario requires. Connecting the fraud-review agent as a callable agent from the intake agent directly satisfies the real-time, low-integration-effort requirement.