Sharing what I learn while building in public
Three real incidents on brandon-ta.com, from invented blog claims to a wiped node_modules, and the written rules that now stop each one.
What Anthropic announced for Claude from July to early October 2026: Opus 5, Fable 5.1, Opus 5.5, Sonnet 5.5, pricing and Claude Code changes.
Where DeepSeek, Qwen, Kimi, GLM and MiniMax stand in early October 2026: releases, licenses, official prices, adoption data, and what it means for small teams.
Protocols, coding agents, computer use, METR time horizons and the open problems of reliability, cost and security, with dated sources.
Where the term came from, what METR, Veracode, CodeRabbit and Georgia Tech measured, and where vibe-coded software holds up or breaks.
Eight products were locked on 4 Sep 2026. The next day, 96 new candidates were scored against the same weighted rubric, and none reached 6.0.
Shopify's own limits for chargeback evidence — 2 MB a file, one file per evidence type — and which evidence answers which dispute reason.
Fees, refunds, disputes, reserves, currency conversion and payout timing separate a Shopify sales total from the deposit that lands in the bank.
What 30 real WooCommerce homepages looked like under an axe-core scan on 2026-09-03: 27 had a serious issue, and broken link names beat missing alt text.
Measured on the production build: about 30 KB for most pages, 5.2 KB of JavaScript, and the one React island that cost every blog post 102.7 KB until it went.
A behind-the-scenes look at building and shipping my first indie AI product: Prompt Formatter.
Discover common prompt-writing pitfalls and the massive potential of mastering this skill by 2025.
No posts found
