Hire Angular Developers

Angular earns its keep on dense, long-lived operational software — the kind with hundreds of screens, complex forms and users who live in it all day. Its opinionated structure is a cost on a small marketing site and an advantage on a system that four different people will maintain over five years, because there is one obvious place for routing, one for state, and one for the dependency graph.

Where it struggles is where teams treat it as a general-purpose UI library and fight the framework rather than using it. The strongest Angular engineers are the ones who have inherited a large application, found the module boundaries somebody else drew badly, and corrected them without a rewrite.

What “at scale” actually means in this framework

The AdBoard portal is a useful reference point because we run it: Angular 19, fully standalone with no NgModules left, 81 components and 36 services, and 103 route definitions split across ten route files rather than one. State is plain services and RxJS — no store library — which is a defensible choice at that size and an indefensible one at three times it.

The framework decision that ages worst is not which state library you picked; it is where you drew the route boundaries, because those determine what can be lazily loaded, what can be permission- gated in one place, and how much of the application a new engineer has to read before changing one screen.

The interceptor is where an application’s behaviour is really decided

Two rules in ours do more for perceived quality than any component.

The first is single-flight token refresh. When a 401 arrives, one refresh runs and every other in-flight request queues on a subject until it resolves, rather than each firing its own refresh. Without that, one expired token produces a burst of parallel refreshes, and whichever loses the race logs the user out mid-task.

The second is that the global error interceptor surfaces only 5xx. Anything in the 4xx range belongs to the component that asked, because it knows the field and the wording. Skipping that rule is how an application ends up showing two toasts for one failed save — the generic one and the useful one.

Permissions have to survive a hard reload

A guard that resolves asynchronously against an API is correct on a click and wrong on a deep link. The user pastes a URL, the guard has no permission set yet, and it either bounces them to the dashboard or lets them through and fails inside the component.

Our permission guard is a route-data check — data: { permission: 'sites.view' } — resolved synchronously against a set rehydrated from local storage, with a redirect built as a UrlTree rather than an imperative navigation. A permission model is only as good as its behaviour on a cold start, and that is the case nobody demos.

The problems the framework does not solve for you

Two from the same codebase are worth stating because they recur everywhere.

A dropdown inside a scrollable table and a card with a hover transform cannot be positioned in place: the overflow clips it and, more subtly, a transformed ancestor becomes the containing block even for position: fixed. The menu is attached to body and re-anchored on scroll and resize — and explicitly removed on destroy, because a manually reparented node is not Angular’s to clean up.

Drag interactions are set up inside runOutsideAngular and coalesced with requestAnimationFrame before a single detectChanges. Left inside the zone, a pointer-move handler triggers change detection across the entire component tree at pointer frequency, which is exactly the kind of sluggishness users describe as “the app feels heavy” and nobody can reproduce in a profiler trace of a single click.

Where Angular is the wrong choice

Public, content-heavy pages that need to rank. An Angular SPA serving the root of a domain delivers an empty app-root to a crawler — zero characters of text without JavaScript — and no LLM crawler executes JavaScript at all. We moved our own product site to a static generator at the root and kept Angular under /app/, which is the arrangement we would recommend to anyone in the same position.

It is also the wrong choice for a small team building something small. The framework’s structure pays back over years and headcount; on a six-week internal tool it is a tax with no return, and an engineer who cannot say that is selling their comfort zone.

What we interview for

The useful question is not which version someone has used. It is what they do with a table of fifty thousand rows and eight filters that has to stay responsive — whether they reach for virtual scrolling and change-detection strategy, or add a spinner and hope. The second answer is common and expensive.

Then two more. How do they cancel an in-flight request when the user types again, and where does the subscription get torn down — a search box without debounceTime, distinctUntilChanged and a teardown is the single most common leak we find in inherited Angular. And what did they do the last time a library they depended on had not moved to the Angular version they needed.