Most things that work in a demo do not work in production. The reasons are mundane and predictable, but they bite every engineer at some point: the dataset was different, the load was higher, the network was less reliable, the failure modes were assumed away, the test environment matched the production environment until it didn’t. Closing that gap is the single most underrated engineering skill, and it’s almost never the part that gets taught.
This monograph collects ten ideas about that gap — the small handful of distinctions that separate “works on my laptop” from “survives in production.” It’s organized as a working checklist rather than a survey paper.
It’s a companion to Ten Ideas Beyond the Code, which covers the wider concerns around the engineering work; this one is narrower and concerns the engineering work itself, specifically at the moment of deployment.
What it covers
Ten ideas organized into four sections, about thirty minutes to read.
§1 — Correctness Across Environments. Reproducibility. Data contracts. Pipeline parity between dev, staging, and prod. The configuration discipline that lets you debug something on Monday that broke on Sunday.
§2 — Performance Under Real Load. What “fast enough” means. Profiling. Load testing. The handful of techniques that catch the performance failures before the customer does.
§3 — Resilience and Operability. Failure modes. Graceful degradation. Observability. The instrumentation that means the on-call engineer at 3 AM can actually figure out what’s happening.
§4 — Trust and Ownership. Who’s responsible when this breaks? Runbooks. Postmortems. The ownership model that distinguishes professional engineering from heroics.
Written for the engineer who has shipped a demo and is now responsible for the production version.
Read it
← Back to Autonomy