AI-901 ยท Common AI workloads
Chat assistant versus agent: what is the difference?
Chat assistants and copilots vs agents: goal-driven behavior, instructions, tools and function calling, memory and knowledge grounding, multi-step planning and human oversight.
Key points
- A chat assistant replies to each prompt from the model's knowledge.
- An agent is given a goal and instructions, then decides which steps to take: it can call tools (functions, search, code), use memory and grounding data, plan several steps, and loop until the goal is met.
- Sensitive actions should keep a human in the loop for approval.
Source: https://learn.microsoft.com/en-us/azure/foundry/agents/overview
Practice questions
What mainly sets an AI agent apart from a plain chat assistant?
- It always produces longer answers
- It works toward a goal by choosing steps and calling tools
- It can only be used with spoken input
- It never needs instructions
Show answer
Answer: B. It works toward a goal by choosing steps and calling tools
An agent works toward a goal by choosing steps and calling tools. A longer reply, a bigger context window or a nicer voice does not make a system agentic.
Which items do you configure when defining a prompt agent in Foundry?
- Instructions
- A model
- Tools
- A container image you build and host yourself
- A new foundation model trained from scratch
Show answer
Answer: A. Instructions, B. A model, C. Tools
A prompt agent is defined by its instructions, its model and its tools. You do not build container images for it (that is a hosted agent) or train a model from scratch.
True or false: when you give an agent instructions, you retrain the model's weights.
Show answer
Answer: False
False. Instructions are part of the agent's configuration and are sent to the model at run time; the model itself is not retrained.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
