Pi

本期播客深度剖析AI编程工具的工程本质：PI智能体以极简设计实现自我修改，揭示‘暗工厂’式代理泛滥导致代码质量滑坡，并强调人类工程师因‘伤疤’驱动的重构不可替代。

Vercel 推出 AI SDK 7，引入 HarnessAgent，支持 Claude Code、Codex 和 Pi 等代理框架，简化模型切换和代理管理。

Running Qwen 3.5-9B (q4_k_s quantized) on an M4 MacBook with 24GB RAM achieves ~40 tokens/sec, supports 128K context and tool use for local development.

Nick Nisi found that by trimming AI agent skills from 95% down to just 5 core roles (implementer, verifier, reviewer, closer, retro), he achieved higher-quality outputs; the key was replacing Claude-native skills with a TypeScript state machine to solve context loss.

Armin Ronacher criticizes poor quality in current open-source issue reports and advocates for a simplified four-step format.

Waza 已经吸收了大量实际项目经验，现在在 Mac 原生应用、CLI 工具和 Rust 方面的能力与 Web 开发相当，支持多种 AI 代理运行时。

Waza toolchain update, supports multiple AI coding assistants, automated publishing process, but low information density.

The article discusses how to build an event-sourced agent framework using stream processors, emphasizing debuggability and extensibility.

该推文仅简要介绍Mario Zner及其项目Pi（OpenClaw核心引擎），提出‘软件应能自我修改’观点，但无技术细节、实证或架构说明。