STAR Apps: A New Generation of Front-End Tooling for Development Workflows

Avatar of Shawn Wang
Shawn Wang on

Product teams from AirBnb and New York Times to Shopify and Artsy (among many others) are converging on a new set of best practices and technologies for building the web apps that their businesses depend on. This trend reflects core principles and solve underlying problems that we may share, so it is worth digging deeper.

Some of that includes:

Naming things is hard, and our industry has struggled to name this new generation of tooling for web apps. The inimitable Orta Theroux calls it an Omakase; I slimmed it down and opted for a simpler backronym pulled from letters in the tooling outlined above: STAR (Design Systems, TypeScript, Apollo, and React).

STAR apps are not “yet another front-end stack.” They involve additional opinions and constraints. As such, STAR apps aren’t necessarily easy, either. They have a learning curve. A solo developer may find STAR apps unnecessarily verbose because they front-load communication overhead. STAR apps are more about product team workflow than they are about any specific technology.

However, we find that companies upon companies are finding this stack to be a worthwhile investment. We should ask why.

Context: From LAMP to MEAN

The LAMP stack was identified in 1998 by Michael Kunze to describe the combination of Linux, Apache, MySQL, and PHP as predominant open source technologies to write a full web server. In this model, all rendering and logic was done on the server side, and the role of JavaScript was extremely limited. To this day, this is the most common website architecture due to the popularity of long established frameworks like WordPress, which powers 30% of the Internet.

In the ensuing 20 years, the growth of the web platform (JavaScript in particular), led to a evolution of the “front end” discipline, as a complement to “back end” server-side concerns. Through a combination of Atwood’s Law and Metcalfe’s predictions on the triumph of the web over native platforms, these efforts culminated in a re-imagining of the monolithic architecture to straddle both front and back ends. One prominent encapsulation of this was the MEAN stack, coined by Val Karpov in 2013, to offer “full-stack” JavaScript alternatives, including MongoDB (for NoSQL data storage), Express and Node (to write web servers), and Angular (for writing reactive user interfaces).

What’s changed

However, in the last five years, multiple trends have chipped away at the MEAN stack and the ideal of the full-stack JavaScript monolith:

  • Instead of every developer writing bespoke endpoints, APIs have become an economy of their own with companies like Stripe, Twilio and Zapier growing purely through the strength of their APIs.
  • The acquisition of Firebase and launch of AWS Lambda in 2014 — and the subsequent serverless revolution — has made the concept of doing your own undifferentiated server management and reliability engineering far less appealing.
  • As for proprietary backends, it was clear that not all backend environments were going to be written in JavaScript, particularly with the continuing strength other language frameworks, including Rails, Laravel and .NET, and emerging languages like Go. Even the creators of Express.js and Node.js wound up abandoning JavaScript development altogether.

This has meant that the product engineer’s stack and primary work has shifted even more toward the front end over what was envisioned by the MEAN stack. Chris has described this as a phenomenon that gives extraordinary powers to front-end developers because of the trend toward front-end tooling for what’s traditionally been considered back-end territory. Front-end engineering has also evolved, mostly by incrementally adding a constraint layer on top of what we already use — adding a design philosophy, types, schemas, and component structure to how we make our apps.

Why all this change? Stop changing things!

The truth is that we now live in a world where product and business needs now have requirements to bring web app (including mobile web) engineering on par with Android, iOS, and desktop native app development, while our disparate web development tools are still woefully inadequate in comparison to those tightly scoped ecosystems. It’s not that there’s anything inherently wrong with older toolsets or that the new ones are perfect. Instead, the changes can be seen as responses to the underlying needs of product teams:

  • Stronger types: Type-checking isn’t a panacea, nor does it replace the need for tests, but it does enable better tooling and increase code confidence. TypeScript and GraphQL do this for clients and APIs, as Chris Toomey of thoughtbot has shown. Lauren Tan of Netflix has taken this idea even further to propose a full end-to-end Strongly Typed Graph.
  • Integrated designer/developer workflows: A reliance on manual code tests and design reviews doesn’t scale. Design systems now are comprehensive documentation on the how and why of reusable components across an organization. Brad Frost has shown how to set up “workshop” and “storefront” environments for a style guide and design system workflow using Gatsby. Design tools, like Sketch and Framer, have even begun to tightly integrate React and design into streamlined workflows. TypeScript and GraphQL both also offer tightly coupled self-documenting features with TSDoc, GraphiQL, and related IDE integrations.
  • Optimized for change: As product teams embrace iterative agile sprints and split testing, it is increasingly important to use flexible paradigms that embrace incremental adjustments. Dan Abramov of the React team calls this “second order” API design — robustness to changing requirements. Design Systems and React make it easy to compose reusable components at breakneck pace, with TypeScript dramatically shortening feedback loops. Adam Neary of Airbnb shows a wonderful example of refactoring and iterating with React and Apollo GraphQL in production.

Note that “product teams” in this article primarily refer to product engineering teams, though it is often the case that product design and product management are co-located or have heavy, frequent input. Engineering workflows must explicitly take them into account as a result.

Remaining frontiers

Believe it or not, I am being descriptive, rather than prescriptive; I’m not recommending that everybody throw out their code and start writing STAR apps. Rather, I am observing and calling out what I see as a trend where great product teams are all converging on this new pattern. And they just may be on to something.

However, I don’t believe the evolution has reached its conclusion. There are still too many important aspects of modern web app development that need broader consensus, which has resulted in a hodgepodge of custom or one-off solutions and checklists. A big one is performance. The average amount of JavaScript shipped on desktop and mobile has doubled in the last five years. All the wonderful web app engineering in the world will be for nothing if the user navigates away before it loads. The traditional solution has been (often hand-rolled) server-side rendering that’s later managed by frameworks like Next.js and After.js. However, this does still require running and managing a server, so static rendering solutions like Gatsby and React-Static have become popular to render apps straight to static markup to be lazily rehydrated later (the last piece of the JAMstack). Progressive Web App technology and patterns help make subsequent loads even faster and serve as a viable alternative to native experiences.

To be continued…

As this story continues to unfold, I believe that a lot more exploration and experimentation needs to happen to smooth the learning curve for more teams to adopt STAR app workflows. In fact, I am learning about it myself in the open at STAR Labs and invite you to tag along. If you have experiences to share or questions to ask, I am all ears.