antsy

The fail-fast operating philosophy: define how work fails before starting it, never wait on a clock or estimate durations, leave nothing hanging, trace everything with OpenTelemetry, test cheapest-first, treat machine memory as shared, and attribute compute and spend to a purpose. Use it whenever you start anything that runs longer than a moment (builds, test suites, servers, renders, training, local models, deploys, CI), whenever you are tempted to sleep, poll, or say "this should take about N minutes", before any paid API or GPU call, before heavy local work on a shared machine, when adding logging or telemetry to code, when deciding how to wait for something, when first doing long-running work in a project with no .antsy.json, and when judging whether compute, money, or a data set is being used well (ML, evals, tests, automations).