Generating code got easy. Making it last didn't.
I'm István Antal. I write about software architecture, agentic coding, and the craft of building things that last — notes from the work, not the hype.
Featured
Latest
8 postsHow LLM reasoning works
ThoughtsYou're billed for every thinking token, including the ones the vendor hides from you, and more thinking doesn't always mean a better answer.
SpecForge
ToolA menu-bar viewer for spec driven development: one badge counts every change still in flight across all your workspaces, and a click opens the tree.
Where the specs live
GuideSpecs are easier to work with when they follow a familiar structure and you can find the changes that need your attention.
UK earnings calculator
ToolUK take-home pay by year, month and week, plus the 62% marginal-rate corridor between £100,000 and £125,140 where a pay rise is worth least.
Story points were always days
ThoughtsStory points began as ideal days times a load factor. Teams still convert them back, velocity is that conversion in public, and throughput beats both.
AI model benchmarks are a suggestion
ThoughtsModel benchmark tables and one-shot demos are vendor-run, Goodharted, and blind to your harness. The only benchmark that predicts your work is your work.
PDF to Markdown on a Mac
GuideConvert PDFs to clean Markdown locally on Apple Silicon with marker, so LLMs can ingest documents efficiently without rasterising every page.
Video transcripts on a Mac
GuideTranscribe video locally on Apple Silicon with whisper.cpp at 5–8x realtime — no API key, no per-minute charges, no audio leaving the laptop.