Switching to Tailwind CSS

Lee Robinson / December 14, 2020

6 min read––– views

My site is an opportunity to try new tech and form opinions on what I enjoy using. I learn best by building real things. After some evaluation, I've landed on the following tech stack:

  • Next.js (Upgraded to v10)
  • Tailwind CSS (Switched from Chakra UI)
  • next-mdx-remote (Switched from next-mdx-enhanced)
  • Deployed with Vercel

I've improved performance, added new features, and cleaned up some code. Let's dive into why I chose this tech stack.

Next.js 10

Next.js 10 introduced the Image Component and Automatic Image Optimization. I've optimized images manually before (using ImageAlpha / ImageOptim) as well as with automated tools (like ImgBot). With Next.js 10, I no longer need to worry about that.

The Image component helps prevent Cumulative Layout Shift (CLS) by defining the width and height ahead of time. No more jumping layouts. It only loads images as they're scrolled into the viewport, keeping page loads fast. Plus, images are served in modern formats like WebP when the browser supports it.

Preventing CLS required a mental model shift. Previously, my images used normal Markdown image syntax and expanded to fill the width of their container (700px). For example:

![Siamese Cat](/cat.png)

I wanted to avoid using the height and width props of next/image, if possible. The Image component exposes different layout props like layout=fill, which led to me believe I could escape defining sizes. In reality, there's no way to avoid layout shift unless you explicitly tell the layout how much space to allocate. Once that clicked in my head, the Image component made sense.

This meant switching from Markdown image syntax (shown above) to using next/image via MDX. Since I had hundreds of Markdown files, I wanted to automate this. I created a Node script using remark that transformed every image and read the dimensions from the file system. For example, the above Markdown image was transformed to:

<Image alt="Siamese Cat" src="/cat.png" height={50} width={50} />

Tailwind CSS

At first glance, Tailwind seems horrible. I've heard this sentiment many times and read Adam's rebuttals. Earlier this year, I had the opportunity to try Tailwind on a project. I didn't hate it. Still, I needed a larger project to form an opinion on the framework.

Well, the verdict is out. I really enjoy Tailwind. There was a learning curve as I tried to translate my existing CSS knowledge to Tailwind specific naming. For example, this vanilla CSS:

display: flex;
flex-direction: row;
align-items: center;

translates to these utility classes with Tailwind:

<div className="flex flex-row items-center" />

The Tailwind classes feel intuitive and their documentation makes it easy. The underlying idea of being bound to a design system made plenty of sense coming from Chakra UI. For example:

import { Flex } from '@chakra-ui/react';

<Flex px={4} py={2} />;

Chakra's spacing system is inspired by Tailwind. 1 spacing unit is equal to 0.25rem, which translates to 4px by default in most browsers. In the example below, we're adding 16px of horizontal padding and 8px of vertical padding.

<div className="px-4 py-2" />

I still enjoy Chakra for larger applications. Having pre-built UI components saves a lot of time. Their documentation has some good notes on the differences between Tailwind. For further reading, check out How Should I Style My React Application?.

Dark Mode

Tailwind 2.0 makes using dark mode painless. Enable the dark variant and prefix your classes with dark:. That's pretty much it.

<div class="bg-white dark:bg-gray-800" />

I was able to keep a dark mode toggle in the navigation thanks to next-themes. Tailwind and next-themes pair well together.

import { useTheme } from 'next-themes';

const ThemeChanger = () => {
  const { theme, setTheme } = useTheme();

  return (
      The current theme is: {theme}
      <button onClick={() => setTheme('light')}>Light Mode</button>
      <button onClick={() => setTheme('dark')}>Dark Mode</button>

I hit a few issues along the way adding dark mode to Tailwind Typography. The documentation felt a bit disjointed from the rest of Tailwind. I would have expected more guidance on pairing the two together.

After digging through GitHub Issues, I was able to get things working. It wasn't obvious I needed dark:prose-dark. I also managed to get syntax highlighting working through some custom styles.


Libraries like emotion and styled-system require runtime JavaScript to compute styles and generate class names. That doesn't mean you can't use CSS-in-JS correctly. But I'd argue using "plain" CSS helps you fall into the pit of success (CSS Modules, Tailwind, etc) more easily.

There is no correct answer. Each has their own tradeoffs. For my simple site, I choose the performance tradeoff. I was able to reduce the number of assets downloaded by 43%.

  • 📦 Before → ~150kb First Load JS
  • 📈 After → ~85kb First Load JS

This change, paired with next/image, has my Vercel Analytics looking incredible. Zero CLS.

Vercel Analytics

I'm also evaluating using Preact in the client-side production build. This prevents needing next-plugin-preact to handle SSR and HMR (source). So far, I haven't noticed any issues with my small site and the results are promising. 39% smaller.

  • 📦 Before → ~85kb First Load JS
  • 📈 After → ~52kb First Load JS


Last year, I switched from using @next/mdx to next-mdx-enhanced mostly for front matter and layout support. These are now both solved problems. Given that next-mdx-enhanced will likely be deprecated soon (see below) I wanted to explore next-mdx-remote.

You probably should be using next-mdx-remote instead of this library. It is ~50% faster, more flexible with content storage, does not induce memory issues at scale, and fits much better with the way that data is intended to flow through Next.js.

With the addition of getStaticPaths / getServerSideProps, you can treat your MDX data like any other data source in Next.js. It also removes the limitation of being bound to file-system based routing for your .mdx files. Since my site has hundreds of MDX files, moving these to a top-level data/ folder made more sense to me.

Still, the MDX experience feels fragmented. I'd like to merge next-mdx-remote with @next/mdx so there are fewer options. nextra (another related project) allows you to use methods like getStaticProps directly inside your MDX files. Maybe there's a path to convergence for all three.


I added absolute Imports and module path aliases to improve the readability of nested imports inside files. Here's a before and after example:

// Before
import db from '../../lib/db-admin';

// After
import db from '@/lib/db-admin';

I also merged my previous project (jamstackfns) directly into this site with a new section called Snippets! These are interesting pieces of code that didn't justify writing an entire blog post. For example, here's a few snippets I've added recently:

Subscribe to the newsletter

Get emails from me about web development, tech, and early access to new articles.

- subscribers – 26 issues