As someone who built their entire platform on Tailwind and has been running gpt-6-astra for the last 5 days straight working on the migration away from it because it leaves an irrecoverable pile of slop when your project becomes large enough... Heed my warning.

It's not that Tailwind itself is bad. It's good. But agents are notorious for re-inventing the wheel without hard constraints that prevent them from doing it - and when I say hard constraints, I don't just mean an AGENTS.md file or a skill they can ignore. As such, I found that design consistency was near impossible and a constant firefight. It worked at first, but when my project got sufficiently large, it became worse and worse.

I initially tried combatting this with a spaghetti-soup of linter rules, but it wasn't very effective to maintain. So I'm biting the bullet and moving on to greener pastures.

▲

mehagar 3 hours ago | parent [-]

What are you switching to?

▲

nullbio 2 hours ago | parent [-]

I'm probably going to catch heat for this, because people (I did too, I won't lie) have a visceral reaction to the Javascript for styling, but... StyleX.

It came on my radar a few months ago when I first started noticing this rot problem. Then recently I saw that it was getting quite hyped and people have been singing its praise for solving this exact problem. So I'm giving it a shot.

Can't say for certain whether it will work out well or not yet, but based on the feedback I've been seeing from others, I'm hopeful.

i.e. https://dev.to/adioof/stylex-won-css-in-js-because-ai-agents... (not my blog).

	▲	psygn89 a few seconds ago \| parent [-]
		For us we went with CSS modules, it just made more sense for us to stay closer to web fundamentals since our styles are used across different tech stacks. From my experience in the last year, AI is quite good at sticking to your css ground rules and gleaning css patterns from other files all while being underpinned by design tokens (can have linters raise raw values to question also).