TIL the "Gauntlet loop" prompt engineering method came from development of the "Claude of Duty" three.js game;
R3F (ReactThreeFiber), Drei, and NextJS (with OpenNext) save time on generated interactive 3d visuals.
I've coded a few apps with ThreeJS recently as well.
First, I upgraded the ThreeJS particles WebGL example by hand for a number of code test iterations. In the months since, I've been working on superfluid quantum gravity; now with normed tensor gaussian splatters from 4th order tensors to 3rd for 3d and then an affine transformation to 2D.
I have an idea for a Snake-like game to demo the gaussian splatters and amplituhedron-enhanced twistor geometry in a game rendering engine ("splatsnake") that could be built with gauntlet loop ai loops instead of test coverage and loops like I've been running.
(2 weeks ago) "Guantlet Loop" by Matt Shumer https://somethingbig.ai/gauntlet-loop .. https://news.ycombinator.com/item?id=49080957
How to prevent wasteful AI loops without a human in the loop to intercede in death spirals that burn token budgets? Not unique to this loop pattern.
There was this from DeepMind a few weeks ago about online rewriting of agent prompt materials:
"Continual Harness: Online Adaptation for Self-Improving Foundation Agents" (2026) https://arxiv.org/abs/2605.09998 .. https://news.ycombinator.com/item?id=48126112
Costed WASM opcodes would probably help prevent waste by enabling optimization of agent loops.
TIL the "Gauntlet loop" prompt engineering method came from development of the "Claude of Duty" three.js game;
R3F (ReactThreeFiber), Drei, and NextJS (with OpenNext) save time on generated interactive 3d visuals.
I've coded a few apps with ThreeJS recently as well.
First, I upgraded the ThreeJS particles WebGL example by hand for a number of code test iterations. In the months since, I've been working on superfluid quantum gravity; now with normed tensor gaussian splatters from 4th order tensors to 3rd for 3d and then an affine transformation to 2D.
I have an idea for a Snake-like game to demo the gaussian splatters and amplituhedron-enhanced twistor geometry in a game rendering engine ("splatsnake") that could be built with gauntlet loop ai loops instead of test coverage and loops like I've been running.