AI makes AI better.

8 picked from 56 candidates · 31 sources read · 56 loops indexed

1. 01
   Loop of the dayPaperarXiv29 Sep 2026Training data

   AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks

   The loop: The AREX-2 agent synthesizes long-horizon improvement trajectories from ML engineering tasks and uses them to train its successor for better performance on MLE-bench.

   AREX-2 demonstrates that long-horizon reflective data synthesized from ML engineering tasks can effectively train agents to become better at self-improvement.

   loop fit 9/10Hongjin Qian, Chaofan Li et al.via arXiv
   
2. 02
   PaperarXiv29 Sep 2026Self-modifying agents

   SelfSearch: Reward-Free Search for Self-Improving Agents

   The loop: An LLM agent modifies its own instructions and tools using records of its own previous attempts, improving its success rate on future tasks.

   loop fit 9/10Jungwoo Yang, Injin Kong et al.via arXiv
   
3. 03
   ProjectGitHub17 Apr 2026Self-modifying agents

   ModernOps888/the-forge

   The loop: Four LLMs compete to evolve and breed code through a JIT compiler judge, creating a closed loop of code improvement that enhances their own capabilities.

   loop fit 9/10ModernOps888via GitHub
   
4. 04
   ReleaseGitHub28 Sep 2026Self-modifying agents

   algorithmicsuperintelligence/openevolve: v0.4.0

   The loop: The OpenEvolve system uses pluggable selection strategies and program validation to evolve code, capturing token usage and enforcing evolution blocks for its own mutations.

   loop fit 8/10codelionvia GitHub releases
   
5. 05
   Blog postPyTorch2 Oct 2026Kernels and compilers

   Building a High-Performance and Portable vLLM Linear Backend with Helion

   The loop: The Helion autotuner uses a DSL to generate high-performance kernels for vLLM, which then runs LLM inference more efficiently for future tasks.

   loop fit 7/10Sean Chen (Red Hat) and Shangdi Yu (PyTorch, Meta Platforms)via PyTorch
   
6. 06
   PaperarXiv29 Sep 2026Self-reward and self-play

   RLTL;DR: Self-improvement by Internalizing Self-generated Feedback

   The loop: A policy model generates its own feedback insights from failed attempts and internalizes them through training, improving its success rate on future rollouts.

   loop fit 9/10Michael Kirchhof, Eleonora Gualdoni et al.via arXiv
   
7. 07
   ProjectGitHub13 Aug 2026Tools and harness

   ZK-Andy/dsh-continual-evolve

   The loop: A continual self-evolution plugin refines the DeepSeek Harness state based on session trajectories, using a benchmark-driven loop to validate and improve its own harness.

   loop fit 8/10ZK-Andyvia GitHub
   
8. 08
   PaperarXiv29 Sep 2026Tools and harness

   Self-Evolving Harness on Multiple Tasks with the Agent as Its Own Optimizer

   The loop: A frozen model acts as a solver to generate run records and then as a proposer to edit its own harness, improving its performance on subsequent tasks.

   loop fit 9/10Qiankai Xuvia arXiv