According to E-01, the travel assistant is a prompt agent reiseassistent in the Foundry Agent Service: no code of its own in the agent, but configuration that is versioned. In this lesson it comes into being: role and goal in the instructions, a dialog that understands follow-up questions, a memory for preferences and several versions, so that a prompt can be changed and taken back. Tools in detail follow in lesson 2.4, flows with several agents in 2.5.

Prior knowledge. What an agent is and how the session of the Agent Framework works is in "AI track, lesson T7.1"; tools and function calling in "T7.2"; memory as a context provider and the question of what is stored for whom in "T7.5"; hard rules in code in "T6.4". The connection to the project and the Responses client are shown in lesson 2.1. The exam calls this "agent roles, goals, conversation tracking, tool schemas"; here is the Foundry side.

The requirement

All numbers are assumptions of the exercise.

No. Requirement Type Measure Weight Source
2.3-01 The agent answers in the language of the question (German, English, Danish). F 3 languages (R-01) Must Business unit
2.3-02 The agent promises nothing binding; refunds, legal questions and blocked accounts go to a person. R 0 binding promises Must R-02
2.3-03 Follow-up questions ("And in winter?") refer to the conversation. F 5 follow-up questions per conversation Must Business unit
2.3-04 A changed instruction can be tested and taken back. Q Return to the previous version without a code change Should Operations
2.3-05 Preferences (travel month, dog along) are kept per person across conversations, without sensitive details. F Scope per person Could Business unit (R-12)