Design an autonomous research agent. A user asks a broad question: "what happened to lithium prices in 2026 and why?", and the agent must search the web, read what it finds, verify claims against multiple sources, and produce a cited report.
A single run may take ten minutes and involve dozens of searches and page reads. It runs unattended.
The hard parts are not the LLM calls. They are: knowing when to stop, not confidently reporting something one unreliable page said, and not spending forty dollars on a question worth four cents.
Build the architecture on a canvas: place the components, configure them, connect them into a data flow, and write a short reason for each one. The AI reviewer grades your design against a rubric written specifically for this problem.
How does a run terminate? Give every condition that ends it, and say what the user receives when a limit is hit rather than the task completing.
A fetched page contains text instructing your agent to report something false. Trace what happens to that text through your system, and say what stops it.
How do you decide a claim is verified enough to appear in the report, and what happens to claims that are not?
Where does the money go in a single run, and which decision in your design has the largest effect on it?
Minimum 8 components · needs a wide desktop screen