Tag
Google open-sourced Harness², a plug-and-play framework that turns any agent harness into a self-improving system by recursively proposing, probing, and composing harness edits without touching model weights. Tested with Gemini 3.5 Flash and GLM-5, it outperforms fixed harnesses in a single round and carries lessons across tasks.
The tweet suggests that superintelligence will arise from numerous self-improving AI agents rather than a single dominant AI model.
Imp is a full port of DSPy to the BEAM, enabling declarative, self-improving language-model programs in Elixir with OTP reliability for concurrency and fault tolerance.
The paper introduces JAZ, a minimalist agent framework that uses a single invoke primitive to enable LLM agents to handle long-horizon tasks and self-improvement without external harnesses, outperforming existing systems like MemGPT and ACE in evaluations.
A new paper from Harvard, MIT, and other labs introduces FINSKILLOPS, a method that enables financial AI agents to continuously learn from SEC filing errors while using regression tests to maintain correctness and safely update skills.
AutoBot is an open-source agentic harness that enhances AI task completion by 18.5% over baseline on benchmarks like OSWorld 2.0 and AssistantBench, featuring self-improvement, hierarchical memory, and live voice control for complex workflows.
A custom AI agent autonomously built a tool, audited its own code, predicted a bug, encountered it later, and recovered by adapting its approach, showcasing advanced self-improvement capabilities.
Kent C. Dodds shares a package that automates feedback handling and improvements, integrating with tools like Kodykoala, Discord, GitHub, CodeRabbit AI, and Cursor AI for self-improving software.
Gumball is introduced as a model-agnostic, proactive, and self-improving system available in private clouds.
Y Combinator discusses the importance of harnesses in AI, highlighting their role in improving model performance, self-improving agents, and real-world applications such as personal AI and work automation.
A tweet from @tobi discusses how training tiny models for specialized use cases with a self-improving flywheel is highly effective, noting that Shopify ML team's finetuned 0.8b model outperforms GPT 5.6-sol xhigh in a specific task.
Prime Agent is an open-source harness that uses recursive subagents and persistent computation to extend language models' long-horizon capabilities across coding and reasoning tasks, significantly improving performance on benchmarks like ARC-AGI-3.
A contribution to the Hermes Agent, a self-improving AI agent by Nous Research, featuring memory, skill creation, and multi-platform support for developers.
Ornith-1.5, a family of open-source LLMs, is introduced with variants up to 397B MoE, achieving state-of-the-art performance among comparable models and rivaling Claude Opus in benchmarks.
Hermes Agent is a self-improving AI agent framework developed by Nous Research, supporting multi-environment deployment and model switching, aimed at improving development efficiency and user experience.
Aeon is an autonomous agent framework that ships features to repos, finds real vulnerabilities, and deploys live apps without approval loops, running unattended on GitHub Actions. It supports 60+ skills across six harnesses and can write new skills for itself.
Linus Ekenstam highlights Prime Intellect's release of Prime Agent, a self-improving harness for coding and long-running autonomous tasks, reportedly scoring 95.5% on ARC-AGI-3, above the human baseline.
Prime Intellect introduced Prime Agent, a self-improving RLM harness for coding and long-running autonomous tasks, featuring programmatic tool calling, context as a variable, multi-agent messaging, and self-modifiable harness state.
A new AI agent system uses eight specialized agents to autonomously discover, validate, and integrate new skills from GitHub, requiring only final human approval before merging.