Security has always felt like entropy. AI vulnerability research may have just flipped the sign. The short version of our argument.
A storage appliance that shipped before AVX2 existed now runs Google's Gemma 4 26B in my basement, no GPU. Here's how we got a modern MoE model onto a CPU older than the instructions its inference engine assumed were there.
Pricing a shared dev server left a bad taste in my mouth: a Raspberry Pi outclasses a month of mid-tier cloud. So I built a 384 GB, 40-thread server out of retired enterprise gear for under $300. Here's the math, and the honest caveats.
A deep-dive, for people who enjoy this sort of thing: how I took a retired HP StoreVirtual appliance, stripped off its locked-down firmware, quieted its data-center fans, and turned it back into a general-purpose server. Less of a tutorial, more of a restoration log.
When a resource gets cheaper to use, history shows we usually consume far more of it, not less. It happened with coal in 1865, and I think it is happening with software now. A look at what Jevons' Paradox suggests about the demand for development in the age of AI coding agents.
It's a marathon, not a sprint. The conventional choice between speed and resiliency only holds in a window smaller than people think. Measured over the life of an application that actually matters, going well is going fast.
Small problems deserve small solutions
A 100-hour Rails upgrade took 10 hours with an AI coding agent. Here's why that order-of-magnitude shift changes the calculus for apps that have been living on borrowed time.
World Backup Day is a good reminder that backups only matter if you can actually restore from them. Here's why we think exercising your restore process is a core part of responsible software ownership.
We use optional analytics services to understand how people find and use this site. These stay off unless you accept. See our privacy and cookie policy.