It's better to burn out
than to fade away.Neil Young · 1979
I've long felt Codex was harder to use than Claude Code. I went through three months of local session logs: turn by turn, the two look almost identical, and when I had Claude review Codex's work, the verdict was 'better code than I expected.' The hard part is what happens after it's done.
Read on →I added a context-usage readout to Claude Code's status line. Manual /compact went from twice in twenty days to thirty-five times, and my usage quota suddenly started lasting. I didn't change any habits on purpose. I just started seeing the number.
Receivables piling up, payables piling up, and the shop booked solid. I sleep fine — it's the wide-awake hours where all of it sits on me at once. This one is written for me: lay the fears out one by one and see which are real.
I tried superset.sh and multica.ai and never figured either of them out — then realised it wasn't laziness: what they sell, I'd already built by hand. The three things that actually saved me time this week used no tooling at all.
From a hand-written bigram to an agent that decides for itself whether to look something up. Four experiments measured out the same line — and three times I nearly read the result backwards.
Moved the whole blog off Next.js and onto TanStack Start. Redid the styling while I was at it — it had been screaming "template" for too long.
