
Today’s issue: Shopify goes full native, lots of benchmarks, and how Cloudflare’s dashboard triggered my gerascophobia.
Welcome to #520.


Adam getting ready for his first day at Shopify
They say off-by-one errors are the hardest problem in computer science and when Tailwind Labs announced that they were joining Shopify I spent a few seconds mourning the loss of the chance to say “we called it”. But for the record we (mostly) called it.
As we outlined in 465, the writing was on the wall for Tailwind when Adam revealed that their revenue had shrunk to the point where they were forced to lay off 75% of their team. At this point, an acquisition seemed inevitable because no one’s getting rich off GitHub Sponsorships. So now that it’s a done deal, did we all win? Let’s find out.
Was this a win for Tailwind?
Yes. We have no idea what the terms of the deal were, but the fact that Tailwind never raised capital means they probably had two objectives for the acquisition.
It’s safe to assume they were able to do both of these things. But the more interesting question to ask is why Shopify would want to own Tailwind and if it’s a win for Shopify.
Was this a win for Shopify?
I’m going to go with yes. Tobi is nothing if not a patron of the arts open source and now he has another prominent project in his collection. But in addition to the altruistic aspects of Shopify stewarding open source projects, their business still sells website templates. And now that design matters more than ever, having a team who has deep expertise in this area makes a lot of sense.
Was this a win for us?
Ok a little narcissistic, but ultimately we only care about this because we’re all Tailwind users. And yes it’s a win for us too. If Tailwind’s development stopped today, we’d mostly be fine. But as browsers continue to ship useful CSS improvements, having a team distilling this dark magic into bite-sized pieces and improving the project’s tooling is something I think we all want.
Bottom line: You could even say this was a win for Canada, successfully merging two proud Canadian companies making this the ever-elusive win-win-win-win.


Letting agents ship code without error monitoring and protected releases
Sentry is hosting a workshop on September 23 on having systems that help your team ship fast without breaking prod. Using a combination of LaunchDarkly’s protected releases and Sentry’s error monitoring, you can automatically rollback releases if there are regressions in performance or other key user metrics.
In this workshop you’ll learn:
Register for this workshop today.

OpenAI claims to have solved Navier-Stokes (a math problem I totally know about), only to realize that the real challenge in academia is deciding who gets the credit.
The React council of elders has blessed us with React 19.3, the first stable release with <ViewTransition /> and <Fragment ref={fragmentRef} />.
Your agent plans 10 steps, then hits a Cloudflare wall. Apify Web Fetch clears the path, handing back clean Markdown instead of stalling at CAPTCHAs. Try it free - code BYTES gets you $25 in credits. [sponsored]
Flavio Copes wrote an article about how he decides which Cloudflare products to use. I’m old enough to remember when Cloudflare only had one product 😭.
Armin Ronacher is asking why models like Astra are so powerful, but don’t seem to make us any more productive.
Salman Ahmed and Samyak Kathane wrote about orchestrating multi-agent AI architectures with Amazon S3 Files featuring an architecture diagram with no less than 10 AWS services. The use case, you ask? Uploading a form.
Want to become the AI expert on your team? Come to Vegas this October and learn specific tools and workflows from top developers to ship better code faster with AI. Register now. [sponsored]
After 7 years of React Native being the future of mobile at Shopify, they are migrating back to native. “Drop the React… it’s cleaner.” - Tobi.
The Motion docs just got a fresh coat of paint. We all still read docs right…right?
ZeroClick lets AI agents buy from you. It gives you an Agent Storefront & lets them pay you via x402, MPP, credit cards and more, so you can convert agent traffic into paying customers. [sponsored]
Luke Parker created openeval, a “typed prompt-plus-LLM-judge” SDK and CLI. If you know what this means, I’m assuming you love benchmarks.
Speaking of benchmarks, Cursor just released CursorBench 4.0, where they rank models based on “ambiguous, multi-file tasks from real Cursor sessions”.