Operating principles

Short statements that survive repetition — each is a compressed argument with its evidence linked. The test for entry on this page: it changes a decision, not just a mood.

Ship safely, not ship rarely

Our job as software engineers now is to engineer systems, including safety controls, that let everyone ship to production — safely.
— Geoffrey Huntley, “access”

The wall-socket framing that sold it (@rockatanescu, captured because the text resonates): pushing to production should not be scarier than plugging a device into a wall socket. Mains electricity is lethal — a century of engineered safety (fuses, grounding, insulated connectors) made universal access safe anyway. Deploy pipelines deserve the same engineering: the danger is real, so the controls are the product, not the gating.

What follows from it:

• Safety belongs in the system (automated gates, rollback, progressive delivery), not in the social layer (approval theater, release windows, hero review bottlenecks). A safety control that lives in a human’s vigilance is a control that fails on a Friday afternoon.
• “Everyone ships” is the design requirement: when the craft is commoditized (ghuntley’s thesis — access, not skill, is the remaining bottleneck), the engineer’s leverage moves from writing everything to engineering the platform everyone else ships through.
• Fear of deploys is a symptom to fix, not a norm to respect: if people are scared, the missing artifact is a safety control, not more caution.
• Kin to Git hooks (pre-push + credential-scan gates — automated, not vigilant), Agent tests (agents shipping through the same gates), Golden repo (the platform-as-product), and the Musk engineering process step 5 — automate last, after the thing being accelerated deserves it.

Sources: clippings/ghuntley-com-access.md (essay) · clippings/x-2105965581232902464.md (resonance tweet)

Related: Performance engineering (elimination-first) · Software factory (the counter-position debate) · Career development