API integration scaffolding
Your team integrates with a lot of third-party APIs, and someone wants an agent that scaffolds each new integration — client code, retries, error handling. The catch: language models happily invent endpoints that don't exist. Design the agent so hallucinated APIs can't reach the codebase: what grounds its generation, what it's forbidden to fabricate, and what a human still owns. How would you verify its output systematically, and why is this worth building versus copying the last integration?
find_ungrounded_calls(spec_endpoints: list[str], generated_calls: list[str]) → list[str][["GET /v1/users","POST /v1/users","GET /v1/orders/{id}"],["GET /v1/users","GET /v1/invoices"]]out["GET /v1/invoices"][["get /v1/users"],["GET /v1/users/"]]out[][[],["GET /health"]]out["GET /health"]Turn the fuzzy goal into a bounded agent task. Say what the agent reads, what it may do on its own versus draft for a human, and what it must never touch. Then earn the trust: how you’d verify it before rollout, and the pitch (the problem, what you built, why it matters).
Agent build: design the agent in plain language (or narrate it). The coach grades the decomposition, the guardrails, the verification plan, and the pitch.