W02 - Rethinking Complex Abstractions

Reconsidering complex abstractions as AI coding capabilities improve.

Over the past decade we built many complex systems in frontend engineering — component libraries, dynamic systems, configuration-driven platforms. All of this rested on one assumption: high-quality code is expensive, developer time is scarce, and human cognitive bandwidth is limited. So we followed the Don't Repeat Yourself (DRY) principle to avoid duplicated work, reduce maintenance costs, and enable building large systems with limited human resources. This can be called a scarcity-economy model.

By contrast, complex human abstractions create obstacles for AI. When an AI faces a highly encapsulated project full of higher-order components (HOCs) and custom hooks, and those black boxes aren’t clearly defined, it often hallucinates. If code is Write Everything Twice (WET) — straightforward, explicit, even redundant — AI handles it better. We can now reasonably hypothesize, roughly, that if intern-level resources were effectively unlimited, past technical approaches might be rethought. This can be called an abundance-economy model.

I happened to see news about the well-known frontend open-source library Tailwind last week; the incident is interesting and worth a closer look.

The issue started with Tailwind CSS’s GitHub repository. A developer submitted a PRsuggesting adding a /llms.txt endpoint to the official site. This is a plain-text standard aimed at large language models (LLMs), designed to help AI more efficiently and accurately digest technical documentation so it can provide more precise code suggestions while coding. From a technical standpoint, it’s an extremely “correct” proposal. In today’s world where AI-assisted programming is standard, embracing LLMs is almost the politically correct move for open-source projects.

However, Tailwind’s author Adam Wathan personally closed the PR on January 6, citing commercial sustainability. If AI can too easily read documentation and return answers directly, developers will stop visiting the official site. A drop in site traffic reduces exposure for paid products like Tailwind UI and directly threatens the company’s revenue.

Adam’s decision drew criticism in the community, with some saying it betrayed open-source ideals. To explain the situation,Adam released a 33-minute recording—a monologue he recorded on his phone during an early morning walk. Their funds only cover six months of expenses, and they recently cut 75% of the engineering team (in reality three out of four people, but each highly capable). If you have time, I recommend listening; you can sense the low pressure and helplessness conveyed through his voice.

A bit of history about Tailwind. Tailwind emerged in 2017 and, I believe, reached its peak glory in 2020. That year Adam Wathan excitedly announced that theirfirst commercial product, Tailwind UI, surpassed $2 million in revenue in under five months. Looking back now, if the frontend industry had a pinnacle year, I think 2020 largely fits that peak.

Before Tailwind, the most popular frontend component library was Bootstrap. Tailwind displaced Bootstrap by addressing CSS maintenance costs. Tailwind’s philosophy is Utility-First: it doesn’t provide preset components but supplies building blocks like p-4, flex, text-red-500. Essentially it’s the victory of “composition over inheritance” in the CSS domain. Bootstrap follows an inheritance mindset with preset components; Tailwind uses a compositional mindset, assembling atomic classes to create new styles.

Finally, on open source. Some in the community say, “In the end there will only be two kinds of open-source software: projects with a deprecation countdown waiting for unpaid volunteers to exhaust themselves, or projects backed by large corporations.”

What we’re seeing now is Tailwind as the former and Bun as the latter. The root cause is the free-rider problem. Before the AI era, although most users didn’t pay, they at least contributed traffic, reputation, and potential hiring leads. In the AI era, agents have become the ultimate free-riders. They extract open-source knowledge with extreme efficiency, convert it into value for commercial services, and open-source maintainers receive nothing in return.

Tailwind at least exposes that the classic business model of exchanging open-source software for attention, and converting attention into revenue, is undergoing transformation.

But the world won’t be limited to just those two options. Take Stack Overflow, which is an example of a platform with an open-source spirit. Also in last week’s news: after ChatGPT appeared, monthly question volume dropped from 300,000 to almost zero, yet through content licensing and enterprise AI product monetization, annual revenue actually doubled to $115 million.