Articles, research, and technical notes — design process, AI and product commentary, and things I'm learning in public.
OpenAI's DevDay 2026 shipped always-on agents called dots, a shared workspace for them, GPT-6.1 Sol at a fifth of Astra's price, a 300-token-per-second speed tier, and its own answer to Jev. Here's what actually launched, what it costs, and where I'd start.
Higgsfield just shipped Layers — an AI image editor that splits one flat image into real, editable layers. Swap the background, the object, even the text, and nothing else moves. Here's what it actually does, what it costs, and where it breaks.
TypeSafe AI came out of stealth with $40M and a model that can't write a sentence. Three days later it was the fastest-adopted model in Vercel's AI Gateway history. Here's what Jev actually does, what the benchmarks leave out, and the four places I'd put it first.
Cindy Zhu's full setup for a custom Claude skill that reverse-engineers viral short-form videos — install the skill, connect Apify so Claude pulls real transcripts and metrics itself, paste a week of saved reels, and get hook, structure, retention and cross-video pattern reports. Includes the complete paste-ready skill source.
A real-time voice companion running entirely on one RTX 5080. Four stages, five models, ~15.5 GB VRAM, 0.70s from your last word to her first sound. Every number here was measured on the machine, not copied from a spec sheet.
这是一篇用于验证 Notion → 网站发布管道的测试文章。
Why I moved my portfolio off Adobe Portfolio onto Astro, Tailwind, and a Notion CMS — the architecture and the trade-offs.
What changes when the product's core behavior is probabilistic: designing for uncertainty, latency, and trust in AI features.
Product research on the robot-vacuum category: mapping the competitive landscape and where a new entrant could still differentiate.