CSS-in-JS is back
Today’s issue: The ultimate parenting hack, dating advice for New Yorkers, and using TypeScript to offend your agent.
Welcome to #516.
The Main Thing
Developers who never gave up on CSS-in-JS
CSS-in-JS is back
There is nothing like the crippling anxiety that comes with being chronically online, but it’s the price I pay, dear reader, to inform you that in the year of our lord 2026, CSS-in-JS is back.
It all started with a blog post from Linear about migrating their styles to StyleX, which in turn revealed that Cursor has also been migrating their codebase from Tailwind to StyleX.
If you’re not familiar, StyleX is a CSS-in-JS library that Meta announced back in 2019, but it was open sourced in 2023, during peak Tailwind, so it never really took off. But the ideas behind StyleX are interesting. It uses a compiler to avoid generating styles at runtime and instead converts JavaScript objects into atomic styles during the build, resulting in tiny bundles.
You might be asking: since Tailwind supposedly “solved” styling for apps, what’s changed? The answer, unsurprisingly, is agents.
How is StyleX better for agents?
-
Compiler-enforced constraints: As the great Lauren Tan once said, agents need constraints. StyleX restricts how styles can be created, composed, and overridden. These guardrails reduce the number of valid approaches an agent can choose, producing more consistent code as a result.
-
Type safety: StyleX gives agents structured properties, typed tokens, and compiler errors—creating a reliable generate, check, and repair loop.
-
Deterministic merging: StyleX resolves conflicts at the property level, with the last style applied always winning. This means agents don’t need to reason about CSS source order or reconstruct precedence from utility-class strings.
Bottom line: 2026 might be the year where we hop back on the CSS-in-JS bandwagon, but are we really okay with needing 8 different compilers to run our apps?
    |
Our Friends
(With Benefits)
AI agents getting ready to wildly misinterpret your data
How AI agent memory systems start lying to you
What’s worse than having your agent look up some data for you and confidently replying with the wrong answer? Fortunately, Jeremy Daly breaks down the root causes of this on the Oracle Developers blog.
In this post you will learn:
- Strategies for updating derived contexts
- Sync strategies for balancing freshness and cost
- How provenance prevents stale context
- When to regenerate, expire, or preserve context
- How to test and rebuild the derived layer
Read the post to learn how to build reliable agent memory systems from first principles.
Cool Bits
-
Poolside wrote about their model factory. Apparently writing Kubernetes is considered blue-collar work now.
-
Lars Faye makes the case that AI coding prevents people from becoming experts, which is why I make my kids write assembly on scratch paper before they’re allowed to have screen time.
-
Learn how Dhrumil Parekh created an automated workflow using Sentry that fixes bugs, pings reviewers, and eliminates duplicate PRs. [sponsored]
-
Rspack released version 2.2, featuring better performance, support for more platforms, and improvements to
import.meta. -
The legend Kirupa wrote an article titled Moving Around a Grid and Respecting Boundaries, which is an excellent resource if you want to learn how computers move objects or if you’re dating in New York.
-
pnpm is now written in Rust as of this week’s release of version 12.0.0, which means it’s faster—but congrats, you also get some breaking changes.
-
“The app feels slow” is not a bug report. EAS Observe measures render and time-to-interactive per route from real users, so you can name the screen instead of the vibe. Free up to 100k events a month. [sponsored]
-
Vercel Labs created
vgpu, a WebGPU shader library designed for agents. The fact that it’s written in TypeScript feels disrespectful, but I can’t tell who should be offended. -
Cloudflare wrote about how they saved 100 terabytes of memory by optimizing their DNS cache. Did I read that right? Memory?
-
Hyperagent lets you build a team of specialist agents with defined jobs, shared context, and clear handoffs. They run on schedules in the cloud, pass work to one another, and bring you in when a decision requires your judgment. Get $100 in free credits here. [sponsored]
-
Flavio Copes wrote a very timely (for this newsletter) deep dive on StyleX.
-
Calvin French-Owen wrote that small models have arrived, and that what people really want is something that’s fast, cheap, and “good enough”. It’s also why I still eat at Taco Bell.