10 parts · 20 chapters

How FAANG-Scale Frontend Teams Actually Work

A frontend organisation of two thousand engineers is a different kind of thing from a team of twenty, in the same way a city is a different kind of thing from a house. The build must be hermetic because nobody can hold the dependency graph in their head; the framework must enforce patterns because nobody can review every component; the data layer must mask what a component may read because nobody can audit every query. Every piece of machinery exists because a human process stopped scaling, and knowing which process tells you whether you need the machinery.

Ten parts: the org shape, monorepos and hermetic builds (Bazel, Buck, Nx at scale), internal frameworks and what they add over React, state and data layers (Relay, GraphQL, RPC, data masking), design-system distribution with codemods and deprecation at scale, experimentation with guardrails and holdouts, performance budgets and release trains, code review and the quality culture around ownership and lint as policy, on-call for a frontend, and the libraries these companies wrote (React, Relay, Recoil, Jest, Flow, Hermes, Fresco, Lexical) and the problem each one answered.

orgs · build systems · internal frameworks · the libraries and whysenior → staff · engineers joining, building, or reasoning about very large frontend organisations
the orgProduct teams, platform groups, infrastructure, release engineering, and the ratios between them.
hermetic buildsBazel and Buck: every input declared, every output cached, the same bytes on every machine.
internal frameworksWhat a company adds over React (routing, data, i18n, logging, flags) and the cost of owning it.
data layersRelay and GraphQL: colocated fragments, normalised caches, data masking, persisted queries; RPC alternatives.
design systems at scaleVersioning across a thousand consumers, codemods as the upgrade mechanism, deprecation with a date.
experiments and trainsExperiment platforms, guardrails, holdouts; budgets in CI, staged rollouts, kill switches.
cultureReadability, OWNERS files, lint as policy, the review bar, and on-call for a client you cannot recall.
the librariesReact, Relay, Recoil, Jest, Flow, Hermes, Fresco, Lexical: the problem each answered and what it cost.
Paired with Architecture and PlatformizationThe Architecture course covered monorepos, design-system distribution and release trains at the scale of one product. This course is the same ideas at the scale of a company, with the machinery and the culture that scale demands. The Platformization course covers the many-country version of the same organisation.