$ cat ai
AI in practice
What it costs, what it breaks, and where it genuinely replaces work. Written for the people who have to run it after the demo is over.
hardware --check
What you actually need to run a model locally
The binding constraint for running a language model locally is VRAM, not raw compute. As a rule of thumb in 2026, a 7-8B model at 4-bit quantisation needs about 6GB of VRAM, a 13-14B model a…
Updated 2026-09-10
eval --tools
Where AI coding tools help, and where they cost you time
AI coding tools reliably save time on well-specified, self-contained work: boilerplate, test scaffolding, format conversion, unfamiliar syntax and first drafts of code you will review anyway…
Updated 2026-09-10