Ever shorter feedback loop

Krzysztof Hasiński's wroclove.rb 2022 talk prepared the day before to fill an empty slot. Defines the feedback loop as any system where the output affects the input for the next iteration, and traces coding feedback loops from 1960s punch cards (days), via 1970s vt100 terminals + ex/vi (hours), 1980s local compilation (minutes), awk-era automated testing (seconds), to modern exploratory programming, live reload and guard testing. Argues that developers who don't run their own code, push to CI without local runs, or rely on staging click-testing are effectively stuck decades in the past. Prescriptions: spend money on faster developer hardware and CI; run stuff locally (mock third-party services, use production for some investigations, push code often behind feature flags, keep setup simple, dockerize dependencies, provide default config, reduce setup steps, trim dependencies, provide seeds, mind Docker image size); speed up tests (grep for sleep, remove loops/shared contexts that add duplicate tests, shrink fixture files, parallelize with parallel_tests or knapsack_pro, aggregate failures for slow tests, use test-prof to hunt Factory Cascades and prefer build over create). Q&A covers Crystal Ball, paperclip's convert-off trick, GitHub self-hosted runners, Hetzner for cheap CI, a 15-minute test suite cut to 40 seconds by removing a cache-clear hook, factory-bot documentation gaps, guard, callback refactoring, and putting MySQL's data directory in RAM to collapse a 10-minute suite to ~1 minute.