Blog

Recording technical learnings and daily insights.

AIAgentMicrosoft Agent FrameworkAgent HarnessC#.NETTools

The Same Idea with Microsoft Agent Framework β€” Building a Review Agent by Just Adding Tools with the Agent Harness

The 'delegate orchestration, just add tools' idea isn't exclusive to the GitHub Copilot SDK. Microsoft Agent Framework's Agent Harness (HarnessAgent) spins up an autonomous agent β€” with planning, todos, context compaction, and tool approval prebuilt β€” in one line, and you just add tools. We implement the same document review agent in C# and discuss how its design differs from the Copilot SDK version.

Conversation UXLLM AgentEvaluationVoice AIDialogue SystemsLLM-as-JudgeHCI

How to Evaluate Conversation UX β€”β€” A MECE Framework for Agent–Human Dialogue Across Structure, Correctness, and Goodness

A systematic framework for evaluating the conversation UX of LLM-backed agents. We classify the evaluation rules MECE across three separable quality axes β€” structure, correctness, and goodness β€” plus a safety constraint axis, and separate what can be measured deterministically from what cannot. We then show how the weighting of the axes shifts between text chat and voice, all the way down to the cognitive science of turn-taking.

AI AgentLLMClaude CodeArchitecturePaper ReadingMCPContext Engineering

Reading the Design Space of a Production AI Agent β€” Lessons from Dissecting Claude Code's 512K Lines of Harness-Centric Architecture

A comprehensive reading of arXiv:2604.14228 'Dive into Claude Code'. We trace the design philosophy that stacks 98.4% deterministic harness around a while-loop, covering safety, context management, extensibility, multi-agent, and persistence β€” and extract practical implications for building AI agent products.