self-improving

Ornith-1.0 is a new family of open-source agentic coding models from deepreinforce-ai, trained with reinforcement learning that jointly optimizes both the solution and the scaffolding. The 35B MoE version achieves state-of-the-art on coding benchmarks and supports efficient single-GPU deployment.

deepreinforce-ai releases Ornith-1.0-35B-GGUF, a state-of-the-art open-source coding agent model that uses self-improving reinforcement learning to jointly optimize scaffold and solution generation, achieving SOTA performance on coding benchmarks.

LangChain's loop engineering playbook replaces prompt engineering with four nested loops - agent, verification, event-driven, and hill-climbing - enabling AI agents to automatically improve themselves over time. The post argues that building self-optimizing loops is now the key competitive advantage, not using better models.

The author explains how to build a self-improving quant trading system using AI loop engineering, where the AI runs loops to prompt, verify, and act autonomously, contrasting with manual prompting.

deepreinforce-ai releases Ornith-1.0, a family of open-source coding agent models achieving state-of-the-art performance on coding benchmarks, available in sizes from 9B to 397B, with a self-improving training framework and MIT license.

Google released a free guide on building, evaluating, and self-improving Agent Skills, including meta-skills.

Researchers at Nvidia Gear Lab achieved a milestone where 8 Codex-AutoResearch agents autonomously controlled a robot fleet to solve a physical world task without human intervention, demonstrating self-improvement.

Warp CEO Zach Lloyd proposed a two-loop method for an AI Agent's Skill to self-evolve from user feedback, using GitHub issue auto-triage as an example. The inner loop processes new issues, while the outer loop collects signals and distills rules. The framework oz-for-oss has been open-sourced.

This paper introduces LLM-as-Environment-Engineer, a framework where LLMs design their own training environments for reinforcement learning in multi-agent reasoning tasks, enabling self-improving training that surpasses larger proprietary models.

The author discusses the need for a fourth governance loop in self-improving AI agent systems to prevent objective drift, proposing periodic human review, withheld benchmarks, and rotating evaluators as practical controls.

A self-improving GUI agent that learns from its mistakes, using a perception stack with YOLO and OCR, calling LLM only when needed, and running locally. It's open source under the MIT license.

Amplitude introduces Wave, a proactive product agent that automates the build-ship-use-learn loop by analyzing data, surfacing opportunities, and tracking outcomes to help teams build self-improving products.

This paper introduces Self-Harness, a new paradigm where LLM-based agents iteratively improve their own operating harness—prompts, tools, and control flow—without human engineers or stronger external agents, achieving significant performance gains across multiple models.

Introducing Ara, a loop that automatically interacts with your coding agent to generate self-improving codebase pull requests without manual prompting.

Recursi is a self-improving vibe coding environment that operates without API fees.

Autocontext is an open-source recursive self-improvement harness that helps AI Agents continuously optimize through iterative execution, evaluation, and knowledge accumulation, generating reusable playbooks, datasets, and even local models. It is suitable for developers building production-grade Agent workflows.

Proposes SCALE, a framework for self-improving web agents using cognitive-aware exploration with three adversarial roles and a graph exploration strategy. Also introduces a large-scale dataset SCALE-20k from real websites, showing significant improvements in MLLM-based web agents.

A self-improving AI framework that simultaneously updates both model weights and task-specific agent architecture via a language-model feedback agent, achieving significant gains across legal classification, GPU optimization, and biological denoising tasks.

YC General Partner @t_blom presented a talk on building self-improving, AI-native companies, emphasizing recursive AI loops and reducing headcount through AI automation.

This paper introduces a new task formulation for CAD generation that incorporates finite element analysis as feedback, along with improved supervision signals like a text-only blueprint schema and multi-view image renderer, leading to better geometric reconstruction on benchmarks.