🧩 Built an MCP that lets Claude talk directly to Blender. It helps you create beautiful 3D scenes using just prompts! Here’s a demo of me creating a “low-poly dragon guarding treasure” scene in just a few sentences👇

Need this as a widget on desktop.

Give me the timeline where models have good enough latency/bandwidth to real-time render a ghiblified web.

it honestly feels like we’ve gone from “let me spend 4wks building a prototype” to “today I’ll build 4 prototypes to see what I like”. crazy times

is ai model fatigue a thing?

I can't believe I tried to make 3.7 Sonnet work for so long, even though it introduces new technical debt with almost every single change. Take any 3.7 generated file, give it to Gemini 2.5 Pro, and all of a sudden, everything works with half the LOCs

someone needs to benchmark information density per token of output in LLMs. o3 responses feel so much more dense than previous models

rip claude code. the economics just don’t make sense. probably true for codex cli too. aider/cline ftw

Went from bullish Anthropic to bearish. Gemini is going to be better price and quality for agentic coding when they fix tool calling. OAI/ChatGPT are the better chatbot. Hope they're cooking something good.

Every code repo should have enough context in it to allow self-modification. Coding as Code.

Claude 4 Opus -> Claude Opus 4 Why??

Claude Code is the new bash/zsh

Claude Code thin wrapper is the new LLM thin wrapper.

GPT5+Codex CLI may bitter-lesson Opus+Claude Code. CC obviously has the better feature set and utility, but that may be irrelevant if GPT5 is as good as these LMArena mystery models indicate.

With the Anthropic rug pull on quotas for Max, I feel the short-mid term value sweet spot will be a Frankensteined together “Claude as orchestrator/coder, falling back to local models as quota limits approach” tool suite.

Claude Code subagents are a clever way to expose a generic “LLM” MCP tool. The Agent file is essentially a set of static configuration parameters (the front matter) and a system prompt (the markdown). This is exposed to the LLM as a tool with a single “user prompt” parameter.

Ouch.

EVERYONE: pls fix search and notifications LITERALLY EVERY PERSON: pls fix search and notifications NIKITA: what if wdym auto translated to "what do you mean"