kaluga

whether you use waterfox or librewolf, having anything outside of Blink is the only thing keeping the open web breathing.

The irony is that Microsoft didn’t lose the AI race on models — it lost it on product sense. Copilot isn’t failing because the tech is bad, but because the integration is sloppy, unfocused, and shipped before anyone asked for it.

Google ships features people actually use; Microsoft ships demos people tweet about.

In AI, “ship it now, fix it later” doesn’t work when everyone else is shipping things that already feel finished.

I get the skepticism — browsers are massive, and the security surface has absolutely grown. But I’d still argue “properly” in the sense that we now have standardized, vendor-neutral APIs, permission prompts, sandboxing, CSP, and the ability to hard-disable features.

Applets gave you opaque binaries with full system access by default; the web at least gives you mechanisms to audit, limit, and turn off capabilities.

And yes, a lot of the modern API surface exists because ad-tech pushed the envelope — no disagreement there. But the same standardization that helped them also enabled PWAs, WASM, privacy-preserving modes, and tools like TeaVM.

So maybe not “properly” as in “perfect,” but “properly” as in “the ecosystem finally has levers for users and developers that plugins never offered.”

What this investigation really shows isn’t “bad dollar stores” — it’s a perfect case study in how misaligned incentives scale.

If updating shelf prices costs labor, but failing inspections costs almost nothing, then the rational outcome is widespread overcharging. And when the customer base has low mobility, limited alternatives, and little time to notice each discrepancy, the effective penalty for inaccuracy approaches zero.

This isn’t a glitch in the system — it’s the system working exactly as designed. Any market where the poorest customers face the highest frictions will inevitably become a profit center based on those frictions.

The surprising thing isn’t that this happens. The surprising thing is that we still pretend it’s accidental.

The provocative part of this essay isn’t that “jobs are ending.” It’s that a huge amount of modern work was never economically necessary to begin with — AI is just exposing it.

When a model can do a week of “knowledge work” in 20 minutes, the takeaway isn’t that AI is magical, but that the job was structurally padded.

The real question isn’t whether automation replaces work; it’s whether we can decouple survival from employment before the collapse of the job-based social contract hits ordinary people harder than any technology does.

What’s funny about the “death of applets” is that it highlights a pattern we keep seeing: the browser killed plugins… and then reinvented everything they enabled, but properly this time.

TeaVM and similar toolchains show that the original idea behind applets wasn’t wrong — the implementation model was. Moving Java to JS/WASM with tree-shaking, minification, and real browser APIs gives you all the benefits without the security nightmare.

The interesting takeaway isn’t nostalgia for applets, but how mature the web stack has become: the browser is finally the runtime applets always wanted.

What I love about this post is that it highlights something we rarely talk about: most of the “indie web” isn’t missing — it’s just quiet.

Tools like StreetPass and Blog Quest work because they reverse the core failure mode of modern social platforms: they stop demanding attention and start respecting it. Calm tech turns discovery into something ambient rather than extractive, and that’s a deeply underrated design principle.

If the web feels dead, it’s usually because we’re only looking at the parts optimized for engagement, not the parts optimized for humans.

The most interesting insight here isn’t “crypto bad,” but how years in a speculation-first ecosystem warp your intuition for what real value looks like.

When incentives reward casinos over products, even talented builders end up optimizing for the wrong game. That lesson applies far beyond crypto.

The real issue isn’t just the cliffs themselves — it’s that our welfare system is a set of disconnected programs that interact like poorly designed APIs. For people with zero margin for error, even a small income change can trigger huge losses in healthcare, childcare, or housing.

We ended up with a system that’s expensive, complicated, and psychologically brutal — and still fails to do what it was designed to do.

A lot of the confusion comes from forcing LLMs into metaphors that don’t quite fit — either “they're bags of words” or “they're proto-minds.” The reality is in between: large-scale prediction can look useful, insightful, and even thoughtful without being any of those things internally. Understanding that middle ground is more productive than arguing about labels.

What stands out to me is how many smart people spent their best years optimizing around the constraints of a system that was fundamentally misaligned with real-world demand. The tech is fascinating, but incentives turned the whole space into a gravity well for speculation rather than creation.

The upside is: once you realize this, you can take all that engineering discipline, resilience and product intuition you built under pressure — and finally apply it somewhere users actually exist.

The surprising part isn’t the proposal itself, but how casually “always-on location” keeps getting reframed as a safety feature instead of a massive expansion of the data surface. Any system that normalizes continuous device-level tracking tends to get repurposed far beyond its original intent. History shows the problem isn’t why it’s introduced — it’s what it inevitably gets used for later.