Blog
Notes on products, interface language, and calm digital aesthetics.
Why massive context windows in LLMs don’t save us from hallucinations and often make the problem worse. We dive into Lost-in-the-Middle, primacy/recency bias, 2025–2026 benchmarks, and what actually works in production RAG systems.
ReadOn March 31, 2026, npm broke twice in one night. Axios - with 100M+ weekly downloads - shipped a RAT trojan for three hours via a phantom dependency. Meanwhile, Anthropic accidentally published the full Claude Code source: 512,000 lines of TypeScript with system prompts and architecture. I break down both incidents: how the attack worked, what leaked, and what to do.
ReadPOSSE - Publish (on your) Own Site, Syndicate Elsewhere. A principle where you publish content on your own site first, then syndicate it to third-party platforms. I break down why it matters, how it works, and why I built my own site.
ReadI launched a static site on Astro with two languages, a blog, and a product landing page. Before writing any content, I figured out SEO and GEO — optimizing not just for search engines but for language models too. Here's what I configured, and why SEO alone isn't enough anymore.
ReadThe third week of public development of MENO — a Telegram bot for task management. Users reached 216, retention holds at 47%, I reworked the bot's interface and received the first request for group functionality from an English-speaking user on dev.to.
ReadAll my work communication happens in Telegram, while classic task managers require manual copying and a separate ritual. I spent a long time looking for a way to organize the chaos right inside the messenger, found nothing suitable — and ended up building my own bot. I’m sharing the journey from problem to solution.
ReadTasks in Telegram appear faster than you can write them down.
Read