ทวีตที่ปักหมุด
I wanted to share something I built over the last few weeks: isometric.nyc is a massive isometric pixel art map of NYC, built with nano banana and coding agents.
I didn't write a single line of code.
English
Andy Coenen
5.4K posts
Andy Coenen
@_coenen
AI controllability @GoogleDeepMind | ex The M Machine | opinions my own
New Jersey, USA เข้าร่วม Ekim 2014
1.4K กำลังติดตาม5.6K ผู้ติดตาม
Andy Coenen รีทวีตแล้ว
I don't think anyone actually knows what the future of Software Engineering looks like with AI Agents. The frontier is moving so fast, and we are learning so much, that anything you think you know you are likely to learn something different tomorrow.
English
This is so awesome - the distance between idea and implementation has never been smaller
Florian Hoenig@rianflo
On a NOT Unbound note, here is a spare-time experiment I always wanted to try out: Making physically plausible and interesting worlds *without* using Perlin noise and *without* using regular grids. Here is how it works:
English
seriously questioning if agents actually need a "loop" now. the agent context and the chat UI are both projections of filesystem state now
so just need something that detects file changes and runs the next agent turn
fsnotify is all you need?
Mario Zechner@badlogicgames
@gregwedow beatiful. now make it more minimal.
English
Andy Coenen รีทวีตแล้ว
if you’d been walking everywhere and one day you got a car, you wouldn’t just get where you were going faster, you would go to way more places
Drew Breunig@dbreunig
Every software engineer I know using coding agents is busier than they’ve ever been. Code is cheaper, but there’s no shortage jobs to be solved.
English
Andy Coenen รีทวีตแล้ว
this is a great, best-of-both-worlds idea
Mike Freedman@michaelfreedman
Introducing TigerFS - a filesystem backed by PostgreSQL, and a filesystem interface to PostgreSQL. Idea is simple: Agents don't need fancy APIs or SDKs, they love the file system. ls, cat, find, grep. Pipelined UNIX tools. So let’s make files transactional and concurrent by backing them with a real database. There are two ways to use it: File-first: Write markdown, organize into directories. Writes are atomic, everything is auto-versioned. Any tool that works with files -- Claude Code, Cursor, grep, emacs -- just works. Multi-agent task coordination is just mv'ing files between todo/doing/done directories. Data-first: Mount any Postgres database and explore it with Unix tools. For large databases, chain filters into paths that push down to SQL: .by/customer_id/123/.order/created_at/.last/10/.export/json. Bulk import/export, no SQL needed, and ships with Claude Code skills. Every file is a real PostgreSQL row. Multiple agents and humans read and write concurrently with full ACID guarantees. The filesystem /is/ the API. Mounts via FUSE on Linux and NFS on macOS, no extra dependencies. Point it at an existing Postgres database, or spin up a free one on Tiger Cloud or Ghost. I built this mostly for agent workflows, but curious what else people would use it for. It's early but the core is solid. Feedback welcome. tigerfs.io
English
Literally a book called “meditations”
Marc Andreessen 🇺🇸@pmarca
Marcus Aurelius, Meditations, X.16: “To stop talking about what the good man is like, and just be one.”
English
Matt Maher tested frontier models in Cursor v. other harnesses. Cursor boosted model performance by 11% on average:
Gemini: 52% → 57%
GPT-5.4: 82% → 88%
Opus: 77% → 93%
His benchmark measures how well models implement a 100-feature PRD. @cursor_ai consistently outperformed.
English
Introducing Texel Splatting: Perspective-Stable 3D Pixel Art
open source paper+code
Most 3D pixel art techniques (e.g. t3ssel8r, ProPixelizer) snap pixels to a screen grid, which only works with an orthographic camera
Texel splatting solves this for perspective cameras: first,
English
should i publish this? basically i used old-style statistics (spectral analysis) to figure out the latent space of a bunch of recipes. Analyzed about 180k recipes and it lets you drill into cuisines and/or features. Also thinking of a seasoning directory to tell you what flavors pair well.
English
@nsthorat Would love some details! I feel like there are certain things that I still want to control with knobs and buttons, but maybe I’m just a dinosaur
English
I built a few major personal projects as apps, and now I’m slowly converting them all to skills. They all work way better now.
Skills are the new apps. Huge mental model shift.
English
@tomduncalf not so much anymore but still get to work with these awesome dudes from time to time!
English
cc @tomduncalf 😎
Jesse Engel@jesseengel
We architected the plugin using JUCE/C++ and React/Typescript - a clever idea presented by @spencersalazar at ADC 2020. This approach ensures latency sensitive operations (audio streaming and processing) happen in a compiled and unmanaged language with a tight clock, while leveraging web technologies for rapid prototyping of the UI (@reactjs, @shadcn, and @tailwindcss). State is synced between Typescript and C++ using Zustand’s state management and nlohmann json. The based frontend also makes it easier for developers to open the codebase in Antigravity (or similar LLM coding tool) and vibecode new interfaces and functionality.
@aidenybai spot on - the most important thing to build is the meta system that enforces code conventions, adds proper testing, and improves context (eg docs, skills, etc)… feel like we almost need new things at the language level to support this
English
imagine vibe coding a piece of software with zero users here
English
So incredible. @nsthorat is one of the best engineers I know and it's incredible what talented people can do with agents working on their behalf. We're about to have so much more cool stuff in the world
Nikhil Thorat@nsthorat
🏎️ gemma-webgpu: a zero-dependency, blazing fast Gemma 1B running entirely in your browser. Full vibe coded from my cell phone. 🔥 136.8 tok/s on M4 Mac (3.3x faster than transformers.js) 📱 101 tok/s on iPhone 17 (270M), 34 tok/s (1B) What we built from scratch: • 18 hand-written WGSL compute shaders with fused ops (fusedNormAdd saves 36 GPU dispatches per forward pass) • Q8_0 dequantization directly on GPU — higher quality than q4 AND faster • Range request streaming loads weights layer-by-layer (~44MB chunks), uploads to GPU, frees JS memory immediately. Peak heap: ~50MB even for the 1GB model • That streaming trick is what makes 1B run on iPhone. it never holds the full model in RAM 12KB gzipped. Zero dependencies. npm install gemma-webgpu
English
