Engineering

The Digital Archaeological Dig I recently assisted a friend with a codebase that felt like a digital archaeological dig—a chaotic mixture of Python versions and conflicting modules. The modern instinct was to upload the lot to an LLM to “fix” everything. The result was absolute carnage. No version control or history meant “editing on master” in real-time. More time spent reversing AI “improvements” than fixing the original bug. A stark reminder: tools are getting smarter, but engineering discipline is becoming a rare commodity. The Return of the Artisanal Mess This mirrors the “artisanal” FrontPage websites of the late 90s. A lower barrier to entry does not guarantee higher quality output. ...

Don’t confuse “work ethic” with “work volume.” Bryan Cantrill’s latest post is a precision strike on the “37k lines of code a day” crowd. In engineering, “virtuous laziness” is about spending hours in a hammock to find the one abstraction that saves weeks of toil. It’s about minimising cognitive load because human time is finite. LLMs have no such constraints. They don’t feel the “cost” of complexity, so they’ll happily bake you a 37,000-line layer cake of garbage just because it’s “free” to generate. ...

Standard serialisation protocols have gone AWOL. If you want to get the packet home with only 50 bytes of airtime, you need to enlist bitpacking.