There's no reason for software to be slow anymore

Dan Luu has been the internet's performance conscience for years, and this piece is another brick through the window of the "it's fine, just add more servers" crowd. The core argument—that hardware has outpaced software by such a margin that most slowdowns are self-inflicted—isn't new, but it's worth repeating until it sticks. He's not talking about exotic HPC tricks; he's talking about basic algorithmic hygiene, avoiding needless allocations, and not bolting on a framework that does ten times the work you actually need. What makes this piece land is the receipts. Luu doesn't just theorize; he points to concrete examples where a few hours of profiling turned a sluggish mess into something snappy. The uncomfortable truth is that most teams don't have those hours because they're buried under the complexity they created. We've built a culture where adding a dependency is easier than thinking, and where "performance" is a four-letter word until the CEO complains about load times. That's not engineering, that's arson. The broader theme here is the collapse of craftsmanship. Open source gave us the building blocks, but it also gave us the excuse to never understand them. Why learn how a database works when you can just add an ORM? Why understand the network stack when you can just add a CDN? Luu's point is that this debt always comes due, and it comes due in the form of janky UIs and 10-second cold starts. The source URL (https://danluu.com/perf-opt/) is a reminder that the fix is usually boring: measure, find the hot spot, fix it. Repeat. I'll go one step further than Luu: this is a licensing and community problem too. We've normalized shipping half-baked performance because the "community" rewards novelty over rigor. A new framework gets stars; a well-optimized library gets ignored. Until we start treating performance as a feature—not a bug fix—we'll keep getting the slow software we deserve. Dan's just the guy holding up the mirror.