Autoresearch: Agents researching on single-GPU nanochat training automatically. Agents autonomously edit and run single-GPU nanochat training experiments overnight, iterating models and logging results via program.md-driven workflows. This shows agentic experimentation loops moving from lab demos to practical, local model iteration — a clear example of Principle 03 and building the island for continuous model improvement.

SWE-CI: Evaluating Agent Capabilities in Maintaining Codebases via Continuous Integration. The paper evaluates LLM agents’ ability to maintain real-world codebases through CI-driven, long-term evolution, prioritizing maintainability over one-shot correctness. Use this as a benchmark and test harness to measure agentic reliability and to design CI gates that validate outcomes — directly relevant to Principle 16 (Validation) and Principle 14 (Immune System).

Guild.ai raises $44M and hits $300M valuation to power enterprise AI agents. Guild.ai is scaling enterprise development, deployment, and observability for AI agents with fresh funding and traction. That commercial momentum translates to mature orchestration and observability primitives you can adopt for agentic coordination and monitoring — see Principle 09 and Principle 14.

Agent Safehouse — macOS-native sandboxing for local agents. Safehouse enforces a kernel-level, deny-first macOS sandbox that prevents local agents from accessing files outside your project. This gives practitioners a concrete pattern for safe local agent deployment and least-privilege defaults, useful for Gate and security hygiene in agent pipelines (Principles 07 and 14).

How to run Qwen 3.5 locally. The guide walks through running Qwen3.5 locally with GGUF quantization and 256K+ context support on low-memory devices. Long-context local inference changes agent architecture decisions — enabling stateful conversations, private data handling, and offline islands of capability that reduce cloud dependency (Principles 07 and 06).