Sacha Greif

Sacha Greif

@sachagreif.com

Founder @ Devographics. I run the State of JS/CSS/etc. surveys. Kyoto, Japan. Into hiking, climbing, making music, LEGO… https://sachagreif.com/ https://devographics.com/

6.4K followers664 following1.2K posts

I'm starting to reconsider my decision to leave X. I'm not sure the current situation where the rather large subset of the tech community with toxic opinions are left to propagate them among themselves is such a great outcome…

It's crazy how normalized racism has become in just a few years. The Great Replacement theory is just acceptable mainstream politics now.

Claude just told me "I’ll read through the data myself first, without using the LLM". Wait… I thought YOU were the LLM…?

Sacha Greif reposted

More range sliders with fancy shapes? Let's go! This time with a tooltip showing the value. curve, wave, zig-zag, arc ... Which one do you prefer? css-tip.com/shaped-range... 0% JavaScript and 100% CSS magic (border-shape, shape(), scroll-driven animations, and more!)

CSS by T. Afif@css-only.dev

Who said range sliders need to be straight and boring?! Hey CSS, distort the line! 〰️ css-tip.com/wiggly-range... Another funny demo powered by border-shape, scroll-driven animations, and more The HTML code? Only an input element (no extra code is needed)

I've been doing "State of…" surveys for 10 years now, and it really feels like it somehow coincided with an era of web development, from the rise of React and Vite to the adoption of TypeScript. In 2026, it feels like we're now at the beginning of something entirely new and different altogether…

At an individual level, AI coding makes many things way faster and easier. Yet many are skeptical that it actually makes companies more productive. Are they wrong? Or are they right, and maybe AI is being misused by large companies?

My theory: going from 1 engineer to 2 is a big difference. But if you already have 1000 engineers and add another thousand, the productivity gains are not going to be as dramatic because of diminishing returns. In an enterprise context, AI coding is probably more like that second scenario?

And I'm guessing the reason why everybody is focused on these huge companies rather than individual makers who can now make apps, music videos, etc. that would've simply been impossible before is that this is where the money is.

View full thread

My conclusion: in 2027, anything you produce must be more engaged, more direct, more impactful. In other words, more human.

View full thread

But none of that addresses my initial concern: at the end of the day, who is this all for? Are we collecting important data for researchers, academics, and regular readers; or are we leveraging AI to produce more and more content that will never be useful–except to be summarized by other AIs?

I'm pretty sure that if I don't course-correct now, within a year my own work on similar surveys will be dismissed as AI slop. In fact, it's already happened in a recent @front-end-fire.com episode, where a mistake *I* made was attributed to AI instead.

View full thread

Don't get me wrong, the report they produced is amazing. And the infrastructure they've developed shares a lot in common (data export, open-source question definitions, etc.) with the concepts I've used in the Devographics surveys myself (and even expands on them).

View full thread

Stack Overflow's solution? *More* AI! They actually have "summarize with Claude/ChatGPT" quick links in this year's survey results.

But if your content needs to be summarized by Claude, why not just feature the summary from the start? Or maybe a summary of the summary, to be even more effective?

View full thread

The problem is that human curation is precisely what gives these reports their value. You invest the time to dig through the data because someone decided you needed to see these charts.

When faced with pages after pages of AI-generated conclusions, it's easy to start wondering what the point even is. And eventually, that attitude can even start contaminating how you feel about *human*-produced reports.

It's now trivial to point LLMs at a dataset and ask them to spit out reports and visualizations that are way more elaborate than what could be reasonably expected from a human.

I'm now at the point in my inevitable cognitive decline journey where I can stare at a new AI startup's homepage for 5 minutes straight and still have no idea what they do

An AI downside I don't see discussed much is how it devalues *non-AI* work. Things that would take weeks such as writing a report, coding an app, or drawing a detailed illustration now seem less impressive because of the possibility they might all be AI-generated–even when they aren't.

But this drives at something else: you can't rely on effort alone anymore to give value to any kind of intellectual output, because that effort has now effectively been reduced to zero. So your thing, whatever it may be, needs to stand on its own even more than before.

A few years back we were all amazed at AI-generated art; whereas today the vast majority of it feels cheap, generic, and uncool. And we detect it pretty easily. I wonder if AI coding will follow a similar trajectory. Will we look back at over-complicated mid-2026 AI slop apps the same way?

so apparently YouTube Music has decided that I'm really into rappers who hate vowels and push the distortion slider to 11, and I'm not sure how I feel about it…

Introducing the State of Devs Correlation Spectrum Here we can see that having no interest in joining a union is moderately correlated with: - Being worried about the erosion of traditional values - Being straight - Feeling curious about tech. Explore the data: 2026.stateofdevs.com/en-US/

These correlations are generated automatically for every single question by the way. It's a great way to explore the data and find new patterns.