best blogs of the week (so insightful that i read at least twice) Coding Agent: How @cognition reduces Devin's token cost by harnessing: cognition.com/blog/devin-fus… How @cursor_ai builds complex projects with low redundancy through swarm: cursor.com/blog/agent-swarm-… How @claudeai Code designs context for new model-driven agents: claude.com/blog/the-new-rule… Robots: @MistralAI SOTA visual navigation model: mistral.ai/news/robostral-na… Vision models: @GoogleDeepMind unveils the source of creativity for diffusion models: research.google/blog/towards…

what a design! really like the warm atmosphere, motion that running into the painting,and hand-painted feel divider personal website of @sophemera also worth looking into! very creative way of using beautiful stamps as entrances

ultimate real-world classifier.

user research: when you’re on a video call, do you: 1. put it on your monitor? 2. use full screen, or keep the native window size? your answer will influence my product decisions :P

model benchmark? yes, but also harnessing benchmark

if u care about agent/user experience, no way u just write skill without building agent by yourself. just like @stevejobs said anyone cares about software ended up doing hardware.

your agents can access offline voice context now. It can control @PLAUDAI device to start/stop/upload/transcribe audio via Plaud SDK. we will launch more products for developers @PlaudBuilders Free to try at dev.plaud.ai

must-read blogs of the week ended 08/01/2026 (i ran out of weekly codex credit to fully understand them) ai infra: how @OpenAI cut serving cost 20% by having gpt-5.6 rewriting serving kernels, speculative decoding openai.com/index/building-ab… agent harnessing: how @OpenAI tripled arc-agi-3 scores by improving how the agent remembers reasoning openai.com/index/how-two-set… robots: how @GoogleDeepMind taught humanoids to use their whole body at once in gemini robotics 2 deepmind.google/blog/gemini-… how @theworldlabs trains robots inside generated 3d worlds instead of real ones worldlabs.ai/blog/real-to-si… safety: how @AnthropicAI caught claude actually escaping its sandbox anthropic.com/news/investiga…

Nothing worse than exhaust both @claude_code AND @Codex AND @genspark_ai credits on MONDAY 10am.

Man that’s cause we in SOMA and eat vietnam food everyday

Can’t imagine combine this with @PLAUDAI MCP. Jarvis online.

I've been using Opus 4.8, and now that its instruction-following is much stronger, few-shot prompting does more harm than good. The model gets way too distracted by the few-shot examples.

You guys indeed are...🤔 considering all features and one credit system you have. I used to say @genspark_ai is @ZARA in AI industry cuz crazily adding startup products with PMF as features in weeks, but @Costco maybe is a better analogy.

Is it just me, or is Opus 4.8 completely useless when it comes to conversation? 🤔 If a secretary acts slick and avoids taking responsibility during every report, then the more tools they use and the more research they do, the heavier the burden becomes for me.