Imo web dev gives a misconception that visuals of game engine - the editor & assets lib - ARE a game engine. In reality 90% of time is spent on per-device bs like iOS memory, sequential fetching, and that fkn magnifier.
Vibing a #threejs engine is easy - shipping with it ain't
"ultra-performant game engine and editor"
"built with threejs, deployed with Electron"
I honestly don't even know how to react to these things anymore...
@bcodecraft https://t.co/FtDFCUmsPo (direct r3f editor)
https://t.co/soXXnFlz3l (a derivative of three, not direct 1:1)
https://t.co/jiIku3XXMk
https://t.co/egHgR6gV8C (SDF + pixel streaming)
https://t.co/xH1t0PNFUN (since three really is just the Scene graph of glTF + scaffolding code)
@endel@cursor_ai Claude 20x is really good atm, fast and high quality code!
$100 (5x) tier is enough for a 1x independent OpenClaw on Hetzner + 1x Claude Code for manual tweaks locally
On $200 (20x) you can run a small personal SWE consulting business π«£
damn we got @KenneyNL vs @mrdoob before GTA 6 ππ
(some three.js AI-related content is relevant, esp. for non-tech people who just want to generate audio demos or small games - but some is obviously just pushing Google genAI because they lowkey have to... π₯²)
whom'stever kept rejecting AI cause "it codes slop" and "it can't understand the entire repo" ...
soon could be a good time to start having that existential crisis π«£π¦ #100ticketsPerHour
@alexoakdev Lock Opus / Sonnet at 4.5 and donβt let it auto-fallback to 4.6
The new β1M context windowβ drains the usage almost instantly - going back to 4.5 brings back the previous efficiency at little to no impact to code quality π«‘
We're building the future of the React Three ecosystem π¨π
Think: first-class WebGPU, LLM-friendly tooling, and a rearchitecture for larger, more complex 3D experiences.
Today we're releasing R3F v10 and Drei v11 alphas thanks to @DennisSmolek.
@makc3d Pretty much π
gltf-transform β
draco β
concat all binary files together β
make byte-offset index β
load index file for all models to IndexDB β
modify GLTFLoader to read byte chunks π
A few server-side upgrades
+ EQS per-tick caching
+ branch conditions re-evaluation only if either EQS or Blackboard changed
+ actors queueing capped at 30ms per server frame
death match bot currently at 8ms, including weapon / ammo checks, reloading, LOS, pathfinding, tracking last enemy position
128 bots at <30ms