AI makes AI better.
A daily hub for work where AI is part of the loop that improves AI, each entry with its loop spelled out. Read more →
8 picked from 13 candidates · 31 sources read · 70 loops indexed
- 01Loop of the day21 Oct 2026AI R&D evals
VeriEvolve: Evaluator-First Evolution for Tool-Using Agents
The loop: VeriEvolve constructs and revises its own executable evaluator using agent run outcomes, which provides a better fitness signal for the next round of agent evolution.
Building reliable fitness signals is a major bottleneck for self-improvement, making VeriEvolve's approach to evaluator-first evolution a critical step forward.
- 0223 Oct 2026Research automation
MLEvolve: A Self-Evolving Framework for Automated Machine Learning Algorithm Discovery
The loop: MLEvolve uses a multi-agent framework to discover machine learning algorithms, which updates its Retrospective Memory and search strategy, making the agent better at subsequent algorithm discovery.
loop fit 9/10Shangheng Du, Xiangchao Yan, Jinxin Shi, Zongsheng Cao, Shiyang Feng, Zichen...via arXiv
- 0321 Oct 2026Self-modifying agents
THUSI-Lab/RoboDoctor
The loop: RoboDoctor performs recursive self-improvement on robot policies, which makes the system better at controlling robots.
- 0421 Oct 2026Self-reward and self-play
AtomSD: Procedural Atom-Guided Self-Distillation for LLM Agents
The loop: AtomSD extracts fine-grained procedural knowledge from a model's own successful trajectories and distills it back into the evolving policy to improve its performance on complex tasks.
loop fit 9/10Bohan Yu, Pengfei Cao, Kai Tian, Chen Han, Shi-Yang Li, Chenxi Zhou, Junlin...via arXiv
- 0521 Oct 2026Tools and harness
Unsupervised Tool Evolution for Coding Agents in Data-Lake Analytics
The loop: TEAL evolves a library of tools and usage skills for a coding agent by analyzing its trajectories, which the agent then uses to solve analytical questions more effectively.
loop fit 9/10Yuqing Yang, Yurong Liu, Deqing Fu, Manav Garg, Sravan Babu Bodapati, Fatma...via arXiv
- 06Tools and harness
HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?
The loop: HarnessDev enables LLMs to design and evolve their own agent harness, which then improves the execution of the LLMs themselves.
loop fit 9/10via Awesome-AI4AI
- 07Research automation
SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness
The loop: Auto-research loops recursively scale to optimize an efficient agent harness, which improves the infrastructure for further agent research.
loop fit 9/10via Awesome-AI4AI
- 08AI R&D evals
EvoHarnessBench: Can Your Agents Keep Pace with an Evolving Harness?
The loop: AI agents are evaluated on their ability to adapt to an evolving harness, which improves the evaluation of agentic R&D systems.
loop fit 8/10via Awesome-AI4AI
Loop index
All 70 entries →
- Gemini models propose code changes
- Evaluators keep best scoring versions
- Finds better Gemini training kernels
- Lowers cost to train next Gemini models
- repeat
- Coding agent edits its own code
- Scored child agents enter the archive
- Archived agents make later self modifications
- repeat
- RL agent lays out TPU blocks
- Better layouts make AI hardware cheaper
- Agent pre-trains on earlier chip generations
- repeat
- Coding agent edits LLM training script
- Runs a 5-minute training job
- Keeps change if validation improves
- Changes transferred to larger models
- repeat
- Research agent writes harness mutation blueprints
- Coding agent implements harness module changes
- Winning harness becomes the new champion
- New champion harness runs further generations
- repeat
- Aligned model generates training data
- Reward model judges and filters it
- Data trains next intermediate instruct model
- Next model generates better alignment data
- repeat
- Model generates candidate responses to prompts
- Model scores them itself as a judge
- Scores become preference pairs for DPO
- Trained model becomes next generator and judge
- repeat
- Claude Code writes its own code changes
- Changes become next Claude Code versions
- Release becomes harness for next development round
- repeat
- LLM agents write and benchmark kernels
- Kernels speed up AI model training
- Session data post-trains smaller kernel models
- repeat
- Coding agent writes fine-tuning code
- Agent post-trains a base LLM
- Trained model scored on target benchmark
- One turn of model training another
- repeat
- Learned optimizers update other learned optimizers
- Training keeps best performing optimizer parameters
- Better optimizers train each other faster
- repeat
- LLM reads reasoning model thoughts
- LLM flags reward hacking behavior
- Signal feeds back into training reward
- Trains more aligned capable agents
- repeat
Support for loop research
How to apply →Grants
Funding for projects that put AI in the loop of making AI better. Rolling, low paperwork, sized to the work.
About grants →Fellowships
Funded time, compute and mentorship for researchers who want to go deep on one loop.
About fellowships →Compute
GPU time and credits for loop experiments, which almost always have to run more than once.
About compute →Have a loop worth building?
Tell us what the AI improves and how the improvement comes back. Applications for grants, fellowships and compute all start on the contact page.