You are building an AI research assistant for a market research firm. An analyst submits a request like "analyse the competitive landscape for enterprise observability tools and produce a 10-page briefing with sources", and the system produces a cited report.
The work genuinely decomposes: searching the web, reading and summarising sources, pulling financial data from an internal database, cross-checking claims, and writing sections. Some of it parallelises; some strictly does not.
A run may take 10 to 30 minutes and cost several dollars. Failures are common: a search returns nothing, a page fails to load, a sub-task produces something unusable, and a failure 25 minutes in that loses everything is unacceptable.
Analysts must be able to see what the system is doing while it works, and be able to intervene.
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 much autonomy do you give the orchestrator, and what does the extreme in either direction cost you?
A run fails 25 minutes in. What exactly survives, and how does it resume?
Eight of ten competitor profiles succeed. What does the analyst get?
Minimum 8 components · needs a wide desktop screen