CSS Coders Are Next to Disappear
The Cycle
Tables. Dreamweaver. Sliced images. The 1990s web was built by people who dragged elements onto a canvas and exported markup.
Then CSS came, and those people were done. The people who could hand-code -- who understood the cascade, the box model, specificity -- became the new specialists. They knew it. They looked at the Dreamweaver crowd and felt the particular confidence of the person who understands what the tool is doing underneath.
That confidence was earned. And then it was not relevant.
The Dreamweaver people were not replaced because they were bad. They were replaced because the implementation layer moved.
What CSS Coders Actually Were
Strip away the craft mythology and ask: what was the job?
A designer produces a mockup. A CSS coder translates that mockup into styles a browser can render. Input: visual intent. Output: code that matches it. The value was in the translation.
Translation layers get replaced. That is what they do. The moment a better translator arrives -- cheaper, faster, available at 3am, requiring no portfolio review -- the old translator is no longer in the conversation.
LLMs are better translators. Not "better with assistance." Better. A designer describes what they want, pastes in a screenshot, points at a reference. Working CSS appears in seconds. The CSS coder was the middleman between design intent and rendered code. That middleman position no longer exists.
Other roles under AI pressure have somewhere to retreat. Illustrators have style, emotional judgment, the "human authenticity" angle. Musicians have live performance, persona, fan connection. CSS coders have what, exactly? "I hand-typed this flexbox"?
The Productivity Framing Trap
Here is the misread doing real damage right now, and it is particularly sharp in Japan:
Companies see Claude Code and think: productivity tool. Their CSS team will now produce three times the output. AI will help them write better code faster.
This is wrong in a structural way, not a quantitative one.
"10x faster" and "why do we have this role" are not on the same axis. One is about throughput. The other is about whether the function exists at all. Treating them as the same question is how you end up surprised.
What actually happens:
Before: 5 CSS coders, output X. The optimistic version: 5 CSS coders + AI, output 3X. What actually happens: 1 designer + AI, output 3X. The 5 are gone.
The lag between "we adopted AI" and "wait, why does this department exist" is one or two years in Japan. The question is already being asked elsewhere. By the time it arrives here, the restructuring has already happened and the job market is flooded with CSS specialists who have nowhere to go.
Low Barrier Was Always the Warning
Some people built CSS careers by learning the skill in a few months and landing well-paid work. The entry barrier was genuinely low. That is what made the path attractive -- no degree, no long apprenticeship, no narrow gatekeeping.
Low barrier to entry is not a feature. It is the signal that tells you how replaceable the skill is.
The arbitrage window was real while it lasted. What took months to learn is now a prompt. The clients who used to call a CSS freelancer are trying the prompt first and calling only if it fails. Mostly it does not fail.
The same logic applies to the corporate employee who quit a stable job to go independent on CSS work. Freelance CSS is especially exposed because there is no organizational complexity to buffer against replacement. Clients pay per output. It is easy to A/B test "let me try AI first." The test keeps passing.
Easy to learn. Easy to automate. Always.
The Exposure Spectrum
Not everyone is equally targeted. There is a rough order:
CSS coder -- maximum exposure. Closed problem. Unambiguous specs. Deterministic output. Frontend developer -- more judgment, more complexity, some buffer. Backend developer -- production chaos, system dependencies, more buffer still. Systems engineer / Architect -- ambiguous requirements, organizational politics, legacy systems with no documentation. Substantial buffer.
The pattern: the more the work involves judgment under ambiguity, the harder the replacement. The more it involves precise translation of clear specifications, the easier.
CSS is a closed problem. There are right answers. "Center this div. Make it responsive. Add a hover effect." No interpretation required. AI handles unambiguous specs extremely well -- that is, almost definitionally, what it is for.
The CSS specialist had no ambiguity to hide in.
Who Survives Each Wave
The people who survive are not the most skilled implementers. They are the people who own the intent -- the designers who know what should exist, the product thinkers who know why, the engineers who understand which problems are worth solving.
The CSS purists who looked down on Dreamweaver users for not understanding the underlying code are now being replaced by people who do not need to understand CSS at all. Same displacement. One level up the abstraction stack.
They were always a translation layer. A more sophisticated one than the Dreamweaver crowd, but a translation layer.
The Dreamweaver users did not see it coming either.
The Cycle Completes
Tables + Dreamweaver: extinct. CSS specialists: endangered. AI-generated CSS: current.
The implementation layer gets eaten. It always gets eaten. The question was never "is this layer safe" but "how far up the stack is the current wave, and am I above or below it."
CSS coders are not unlucky. The role was structurally targeted. The traits that made it attractive were exactly the traits that made it automatable. The Dreamweaver parallel was always there.
There is no CSS equivalent to pivot to this time. The Dreamweaver crowd could learn to hand-code. The CSS specialists have nowhere to go that the prompt cannot follow.