Two Queues, One Cron Tick: Serverless Batch Pipelines

Daniel's episode-generation binge kept triggering redundant site builds, and his attempt to fix it with a polling queue produced the project's only surprise bill. This episode works through the professional version of the pattern: capture stays ad-hoc, processing becomes batched, and the queue lives in a database instead of a long-running worker. We cover why polling is the canonical serverless anti-pattern, how a status column plus a scheduled batch job replaces a worker entirely, and why the two-queue split (generation, then deployment) makes human review a first-class stage instead of a bolt-on. Then we get into the money: batch inference APIs now offering roughly 50% off in exchange for relaxed latency, where Modal sits in that landscape, and why "never time-critical" is exactly the workload batch pricing was built for. Episode #894497 — open it directly at myweirdprompts.com/894497