Error budget burn response
Design an agent that watches error budgets across your services and acts on burn rate. The controversial part: should it be able to freeze releases on its own, or only recommend a freeze? Take a position and defend it. Cover what it monitors, how it avoids crying wolf on noisy short windows, and how you'd validate its judgment against historical incidents. Then convince a room of product managers that this agent is good for them, not just for reliability.
burn_rate_decision(samples: list[float], fast_window: int, slow_window: int, fast_threshold: float, slow_threshold: float) → str[[6,8,10,12,14,16],2,6,6,3]out"recommend_freeze"[[0.5,0.5,0.5,0.5,9,11],2,6,6,3]out"recommend_freeze"[[0.5,0.75,0.5,1,0.5,0.25],2,6,6,3]out"no_action"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.