AI 编程笔记，第 2 页

DeepSeek Harness separates models, tools, sessions, permissions, and interfaces into plugins that builders can replace and combine.

A hands-on look at DeepSeek V4 Flash 0731, its vendor-reported agent benchmarks, and the coding tasks where I would use it.

CodexAug 23, 2026

A practical decision guide for choosing a GPT-5.6 model and reasoning effort level based on task difficulty, speed, and iteration cost.

LoopX keeps goals, evidence, ownership, quotas, and handoffs stable across long-running agent sessions without replacing the agent runtime.

Claude CodeAug 23, 2026

Practical Claude Code habits that reduce context noise, protect prompt cache reuse, and keep recurring work in smaller sessions.

Use ModLens to route screenshots and image files through a configured vision engine, then return structured evidence to a text-only coding agent.

Install an Unsloth GGUF build of Qwen3.8-27B with Ollama or llama.cpp, then choose a quant that leaves enough unified memory for the runtime.

Same model, different inference stack, wildly different latency and cost. Modular's handbook covers VRAM math, KV cache, batching, and serving.

CodexJul 26, 2026

A practical look at ChatGPT Voice and multi-folder projects in Codex, including their workflow benefits, limits, and setup details.

A hands-on Kimi K3 backend audit across 68K lines of FastAPI code, including query fixes, measured results, and the full task cost.