People scenarios, delivery judgement, hiring calibration, and retained technical credibility across a mixed loop.
Treat it as a process gap, not a grudge. Find out where and when decisions are made, get your team's constraints in front of the deciders before the meeting through short written proposals with a comment window both time zones can reach, and escalate a specific costly decision, not the people who made it.
Evaluate feature flags in an SDK from a cached ruleset, assign users by hashing a per-experiment salt with the user id so assignment is sticky, log exposure where the arms first diverge, gate readouts on an automated SRM check, and make the kill switch a top-precedence rule that works with the control plane down.
A notification system at 50 million a day is sized by bursts and provider limits, not its average of about 580 a second. Accept with an idempotency key, apply preferences once, fan out to per-channel queues with transactional and bulk lanes, rate-limit each provider, and never fail over blindly on a timeout.
Describe your use of AI tools by task, each with its check: reading cited lines, running code, confirming APIs against the docs for your version. Refuse any answer you cannot check cheaply or that is costly to get wrong, keep restricted data out of unapproved tools, and back it with one real catch.
Rate limiting an LLM gateway by request count fails because cost is tokens, and output tokens are unknown at admission. Budget tokens per tenant, reserve input plus expected output, settle against reported usage when the stream ends, cap concurrent streams, and cancel upstream on disconnect.
A distributed job scheduler cannot promise exactly-once execution when workers die mid-job, so it promises at-least-once with idempotent jobs. Materialise each firing as a run row, claim it with a lease and heartbeats, judge time by one clock, and key side effects on job id plus scheduled time.