I always found the idea of a newsletter intriguing, it requires quite some effort to come up with something interesting every week or every few weeks. So we decided to do a leap of faith and see what happens.

So here we are, a (hopefully) periodic tech newsletter, completely written with the help of Human Intelligence. We’ll be talking all about AI, what happens in tech, open-source, and maybe we also manage to inspire people to do amazing things.

I don’t know about you, but I really like building software. I’m talking about actually sitting down and exploring the problem at hand, designing and writing the necessary code to make it work, look, and feel right. I love obsessing over the small details — adding a smooth transition between states, neatly aligning elements, adding keyboard shortcuts for those power-users, you get the idea.

I’m not trying to say LLM’s are useless, but they sure as hell lack taste. When you’re doing something you love, you’ll always think ahead ten steps. You ask yourself all sorts of questions — how will it look, how can I make it accessible, how can I make it fast, how can I make it extensible — all of which help guide you when designing and implementing. An LLM won’t, it just does (mostly) what it’s told to do, and even when you give it concise instructions it sometimes fails to apply them consistently.

An LLM doesn’t have eyes to distinguish great UI from bad UI, it doesn’t have an opinion or empathy, and it doesn’t learn after a mistake — after all it’s just a prediction model.

We’re all complaining about how slow Jira is, how poorly optimized every new software or video game is, or how bad Windows (and MacOS) search is, yet instead of focusing our efforts on quality and building truly delightful software we’d rather “ship fast” and look at metrics.

Many people like to praise Apple for how well-polished their ecosystem is — though questionable nowadays — and it’s mostly due to the right people having good taste and putting in all the effort to make sure everyone gets it. Good taste is caring about the experience, it’s about obsessing over the smallest details, it’s about knowing where to focus your efforts to deliver a better product, it’s about knowing what feels and looks just right, but above all: it’s deeply human.

I’d argue not many people have good taste, in fact, most don’t. However, I do think it’s something that can be learned. Finding a great product example is hard, but when you find it you can immediately feel it is different. Building good taste requires analyzing what makes a specific experience feel good, paying attention to each and every aspect of the product and reflecting on how you can apply these to your work. It takes practice, and it requires reflecting over our own mistakes — skills that generative AI will never be able to do.

I don’t really care about the slop that LLM’s generate, and as Linus Torvalds also said, it’s fine for things you don’t care much about, like experimenting with something quickly and throwing it away (definitely not writing a newsletter). What I do care about is our craft and the experiences we build for other humans, which I think is far more important than any amount of profit.

If you don't know what this is, just know that I've spent a lot of time for you to hopefully read this. We're trying something new, so that's why you're receiving this email. You can ignore it, or keep on reading because I think it might be interesting even if you're not too technical.

I always found the idea of a newsletter intriguing, it requires quite some effort to come up with something interesting every week or every few weeks. So we decided do a leap of faith and see what happens.

So here we are, a (hopefully) periodic tech newsletter, written exclusively by Human Intelligence. We’ll be talking all about AI, what happens in tech, open-source, and maybe we also manage to inspire people to do amazing things.

Is slop really the future? I don’t know about you, but I really like building software. I’m talking about actually sitting down and exploring the problem at hand, designing and writing the necessary code to make it work, look, and feel right. I love obsessing over the small details — adding a smooth transition between states, neatly aligning elements, adding keyboard shortcuts for those power-users, you get the idea.

I’m not trying to say LLM’s are useless, but they sure as hell lack taste. When you’re doing something you love, you’ll always think ahead ten steps. You ask yourself all sorts of questions — how will it look, how can I make it accessible, how can I make it fast, how can I make it extensible — all of which help guide you when designing and implementing. An LLM won’t, it just does (mostly) what it’s told to do, and even when you give it concise instructions it sometimes fails to apply them consistently.

An LLM doesn’t have eyes to distinguish great UI from bad UI, it doesn’t have an opinion or empathy, and it doesn’t learn after a mistake — after all it’s just a prediction model.

We’re all complaining about how slow Jira is, how poorly optimized every new software or video game is, or how bad Windows (and MacOS) search is, yet instead of focusing our efforts on quality and building truly delightful software we’d rather “ship fast” and look at metrics.

Many people like to praise Apple for how well-polished their ecosystem is — though questionable nowadays — and it’s mostly due to the right people having good taste and putting in all the effort to make sure everyone gets it. Good taste is caring about the experience, it’s about obsessing over the smallest details, it’s about knowing where to focus your efforts to deliver a better product, it’s about knowing what feels and looks just right, but above all: it’s deeply human.

I’d argue not many people have good taste, in fact, most don’t. However, I do think it’s something that can be learned. Finding a great product example is hard, but when you find it you can immediately feel it is different. Building good taste requires analyzing what makes a specific experience feel good, paying attention to each and every aspect of the product and reflecting on how you can apply these to your work. It takes practice, and it requires reflecting over our own mistakes — skills that generative AI will never be able to do.

I don’t really care about the slop that LLM’s generate, and as Linus Torvalds also said, it’s fine for things you don’t care much about, like experimenting with something quickly and throwing it away (definitely not writing a newsletter). What I do care about is our craft and the experiences we build for other humans, which I think is far more important than any amount of profit.

This couldn’t be a newsletter without also talking about recent developments in the industry:

• Donut Labs, the company that recently shook the markets with their alleged ground-breaking solid-state battery, just shared the first scientific test results conducted by VTT, a highly regarded research institute in Finland, to prove their new batteries actually work. If this technology is actually real, I’m sure glad it’s from Europe. You can follow the progress on idonutbelieve.com.
• With AI completely swallowing all demand for RAM, SSDs and even HDDs, the only obvious way forward is to start optimizing our software (huh, what a concept). Node.js is exploring pointer compression and we might see a reduction in up to half in specific use-cases. Even cloud infrastructure providers are hit by this, I just received an email from Hetzner that they are raising our prices by about 30-40%.
• AI agents are now sentient? Nah, but someone used OpenClaw to shame an open-source contributor into accepting their slop pull-request to one of the biggest projects on GitHub — spoiler alert, they didn’t, but wrote a blog post about it.

Finally, let’s talk open-source. Our whole digital lives rely on open source, so I want to make sure the thankless work of all these people doesn’t go unnoticed.

• The TanStack library family now has a new member: Hotkeys. Now there’s really no excuse to not adding keyboard shortcuts to your apps.
• My girlfriend and colleagues are fed up with this, but if you like maps, I’d highly recommend contributing to OpenStreetMap, you’d be surprise how much more detail there is compared Google Maps. I’ve been obsessed with OSM since we worked on a project that had us deploy our own map solution. I might write about this at some point.
• Offline translation is about to get a lot more interesting. With the recent support added for WebGPU, the community managed to get an ML model working completely locally for translating text with Google's TranslateGemma model. It's not very fast yet, but it shows pretty good results from my testing.

If you read all of that, I like you. It's fine if you didn't, but I'd appreciate if you shared this with your friends, who knows what next topic I'll cover?