Third panel question. Caio: Honeycomb, CloudWatch, pganalyze for DB insights, Apollo for GraphQL, Grafana. Maciej: OpenTelemetry (vendor-neutral — swap backends without code changes) plus past experience with Datadog and New Relic (costly per blog posts); for a GraphQL project his team added custom client-side instrumentation logging request times + params so they could reproduce the exact slow request locally, making the feedback loop tight. Stephen: for smaller/less meaningful apps you can't afford Datadog — he's building a gem that subscribes to ActiveSupport Notifications, pipes them into a SQLite database, and exposes them via a route. Sidekiq ships a good insights interface out of the box. Maciej notes if you already have Prometheus+Grafana from infra, reuse it rather than publishing custom endpoints.