Design an agent with access to 500 internal APIs: CRM, billing, HR, deploys, analytics, ticketing: that an employee can ask to do things in natural language.
500 tool definitions do not fit in a context window, and even if they did, tool selection accuracy collapses long before that. The problem is not calling tools; it is choosing the right one out of 500, and not letting the agent call one the user is not allowed to call.
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 the agent choose among 500 tools? Say what is indexed and how candidates are narrowed.
Where are per-user permissions enforced, and why is filtering the retrieval candidates not sufficient on its own?
Step 3 of a 5-step chain fails. What happens to the effects of steps 1 and 2, and what does the user see?
A tool's schema changes without notice. What in your design contains the damage, and how do you detect it?
Minimum 9 components · needs a wide desktop screen