This code is intentionally ordinary Python, not a live generative model. That’s the point: you learn the control flow and failure boundaries before granting an AI access to tools. Copy it into a local Python environment or read through it line by line.

tasks = [
    {"title": "Review fictional delivery list", "urgent": True},
    {"title": "Draft tomorrow's meeting agenda", "urgent": False},
    {"title": "Check sample discrepancy report", "urgent": True},
]

def triage(items):
    if not isinstance(items, list):
        raise TypeError("Expected a list")
    checked = []
    for item in items:
        if not isinstance(item, dict):
            raise TypeError("Task must be an object")
        title = item.get("title")
        urgent = item.get("urgent")
        if not isinstance(title, str) or not isinstance(urgent, bool):
            raise ValueError("Invalid task format")
        checked.append((0 if urgent else 1, title))
    return [title for _, title in sorted(checked)]

for title in triage(tasks):
    print("REVIEW:", title)

print("No tasks were changed or sent.")

The loop validates input, applies a rule, prints recommendations and stops. It has no network access, credentials or external side effects. A real AI agent adds model reasoning and possibly tool calls, which means more uncertainty and a greater need for tests.