Reading routes from Mr. Guo
AI Agent Engineering and Coding
Read about micro-agents, Claude Code internals, memory and retrieval, AI coding workflows, and operational costs and safety.
A model answering one question and an agent working continuously inside a real codebase require different engineering. This topic connects task decomposition and state control with tools, context, permissions, verification, and recovery. The goal is to identify parts of an agent that can be observed, tested, and maintained.
Start with micro-agent architecture and system engineering for the overall picture, then move into the Claude Code runtime series. Use context and retrieval guides for memory or prompting problems, coding workflow articles for application delivery, and operations guides for tokens, SSD writes, hardware, APIs, and hosting bills.
These pages include source-code interpretations, usage records, and engineering retrospectives from the author and technical collaborators, alongside building guides and reference handbooks. Product behavior and costs change with versions. Keep each article’s tested environment and version assumptions in view; an observation on one machine does not describe every installation.
AI Agent System Architecture Guides and Resources
Understand composable agents through micro-agent design, tool collaboration, hybrid models, and building guides.
Define task, state, and tool boundaries before deciding whether multiple agents are needed. Micro-agent guides, OpenCode, Opal, AgentKit, and small-model discussions provide design references alongside building handbooks. Compare lifecycle control, error handling, and testability rather than assuming stronger models or more agents remove the need for engineering boundaries.
Open this route’s guide →- Article Loop Engineering Explained Through Music: Hooks and Skills Loop Engineering explained through music: understand AI agent loops, Hooks, Skills, Harness, and recurring automation through a musician’s experience.
- Article OpenClaw is just taking off, and now Hermes is here Digital Strategy Review | 2026. Just as OpenClaw begins to gain traction, Hermes arrives to disrupt the market. Is this the next big shift in AI tools?
- Article If OpenClaw's viral success leaves you cold, you're not an average user Unmoved by OpenClaw's hype? You've likely moved past the novelty phase, focusing on core workflow challenges and AI agent deployment rather than just setup.
- Article Decoding OpenCode for Agent Development: The Architecture of Next-Gen AI Coding Assistants A deep dive into OpenCode's architecture, revealing how this next-gen AI coding assistant handles complex programming tasks through session management, tool systems, and provider abstraction.
- Resource Agent System Engineering: From Design to Deployment A deep dive into AI Agent system engineering, covering architecture, reliability, monitoring, evaluation, and key engineering practices.
- Resource AI Agents Companion V2 A comprehensive reference handbook on AI Agents, covering definitions, classifications, use cases, and future trends in the Agent ecosystem.
- Resource A Practical Guide to Building AI Agents A systematic guide to building AI Agents from scratch, covering architecture design, engineering implementation, and best practices.
- Article Deep Dive Into Google Opal: The LEGO of the Agent Era? By Mr. Guo Hey, it’s Mr. Guo. I recently got my hands on Opal, an experimental tool from Google Labs. My gut reaction? It turns something like...
- Article Deep Dive into Google Opal: The LEGO of the Agent Era? A deep dive into Google Opal — how this 'LEGO' of the Agent era makes AI workflows accessible to everyone and becomes a powerful tool for indie developers to rapidly validate product ideas.
- Article A Strategic Guide to OpenAI AgentKit for Agent Builders This strategic guide deconstructs OpenAI AgentKit's core components, strategic intent, and transformative impact on AI agent construction.
- Article The AI Agent Gold Rush Truth: Why 2025 Belongs to 'Super Tools,' Not 'All-Purpose Butlers' The 2025 AI Agent victory belongs to focused 'super tools' not 'all-purpose butlers' — overcoming three limits: mathematical, economic, and engineering.
- Article AI Agent First Principles: Why Claude Code's Minimalism Beats Complex Multi-Agent Systems Explore AI Agent first principles — understand how Claude Code's minimalism outperforms complex multi-Agent systems through simple loops, LLM Search, and structured communication.
- Article Deep Dive: NVIDIA's Bold Claim That AI Agent's Future Belongs to Small Models (SLM) An in-depth analysis of NVIDIA's paper revealing why cost-efficient SLMs outperform LLMs for AI Agents, plus a practical LLM+SLM hybrid architecture roadmap.
- Article The Engineering Path for AI Agents Series (Part 3): From Islands to Federation—Agent Collaboration and Robustness Learn how to build isolated AI Agents into a collaborative intelligent federation, ensuring robust error handling and efficient external interactions.
- Article The Engineering Path for AI Agents Series (Part 2): Building the Agent's Brain and Nervous System Revealing how to build a Micro Agent's brain (decision core) and nervous system (perception/control), mastering prompts, context, and control flow for Agent engineering.
- Article The Engineering Path for AI Agents Series (Part 1): Is AI Agent Dead? Long Live 'Micro Agent'? General-purpose AI Agents easily get lost; the future belongs to focused 'Micro Agents' embedded in deterministic workflows.
Claude Code Runtime Architecture Guides and Resources
Follow Claude Code source analysis through startup, the main loop, task entities, concurrency, permissions, and verification.
Read the source series from startup assembly to the main loop, tool orchestration, agent runtime, and REPL control plane before exploring individual design highlights. Task entities, semantic concurrency, centralized permissions, and worktree isolation explain long-running execution. Some legacy URLs contain different published versions of a topic; follow one version through the series rather than treating repeats as separate evidence.
Open this route’s guide →- Article Claude Code Deep Dive 09: Why Worktree and Remote Isolation are Runtime Semantics Explore how Claude Code elevates Worktree and Remote isolation to native runtime semantics, ensuring secure and efficient AI agent collaboration in development.
- Article 10 Highlights of Claude Code: The Verification Agent's Adversarial Approach Explore how Claude Code's Verification Agent moves beyond passive review, using adversarial testing to force evidence of execution and ensure code quality.
- Article Claude Code Highlights 01: Agents as Tasks An in-depth look at Claude Code's architecture. By treating agents as manageable tasks, it enables background execution, lifecycle control, and robust workflows.
- Article Claude Code Source Code Guide 03: Agent Runtime A deep dive into the Agent Runtime architecture in Claude Code, exploring how unified task models enable core agent scheduling, isolation, and recovery.
- Article Claude Code Highlights 06: Agent Runtime Spec A deep dive into Claude Code's agent definition mechanism. Discover how runtime specs replace simple prompts for precise tool control and system scalability.
- Article Claude Code Highlights 07: Agent-Specific MCP Claude Code uses agent-specific MCP to mount external capabilities by role, reducing tool noise and boosting efficiency and security in multi-agent systems.
- Article Claude Code Architecture: A Multi-Agent System in Your Terminal Claude Code is more than a CLI tool; it is a sophisticated distributed multi-agent runtime designed for task orchestration, isolation, and collaboration.
- Article Claude Code Highlights 03: Fork and Prompt Cache Explore Claude Code's Fork and Prompt Cache mechanisms. Learn how to optimize multi-agent reasoning costs and achieve high-efficiency engineering innovation.
- Article Claude Code Highlights 03: Fork Subagent and Prompt Cache – Rare Cost-Level Innovation Explore the design of Claude Code's Fork Subagent and Prompt Cache, revealing how these innovations drive efficient multi-agent collaboration and cost optimization.
- Article Harness Engineering Deep Dive: How Claude Code Integrates Guardrails into Runtime Explore how Claude Code integrates context management, tool orchestration, and task recovery directly into the runtime to build reliable, production-grade AI agents.
- Article Claude Code Highlights 04: In-Process Teammates Explore the in-process teammate architecture of Claude Code. We analyze how it balances performance and isolation for efficient, controlled multi-agent workflows.
- Article Claude Code Highlights 04: The Teammate Runtime for Balanced Performance and Isolation Explore Claude Code's in-process Teammate Runtime. We analyze how this design balances high-performance execution with secure isolation for multi-agent workflows.
- Article Claude Code Top 10 Features 05: Leader Permission Control Discover how Claude Code uses Leader permission control to secure multi-agent collaboration, centralizing approvals while streamlining complex workflows.
- Article Claude Code Source Deep Dive 04: REPL and AppState Explore the Claude Code source code to uncover how REPL and AppState form a unified Agent runtime, revealing the architecture behind multi-agent interaction.
- Article Claude Code Source Deep Dive 04: REPL and Control Plane A deep dive into Claude Code's REPL and AppStateStore. Discover how these components serve as a unified control plane for complex multi-agent runtime management.
- Article Claude Code Top 10 Features 06: Runtime Spec Explore Claude Code's Runtime Spec mechanism. Learn why agent definitions should be runtime specifications rather than prompts to build robust AI systems.
- Article Claude Code Highlights 02: Semantic Tool Concurrency Explore Claude Code's semantic tool concurrency. Learn how dynamic runtime scheduling balances execution speed and state safety for stable AI agent workflows.
- Article Deep Dive into Claude Code: Transforming Terminal Assistants into Multi-Agent Runtimes A deep dive into Claude Code's architecture reveals its core: a sophisticated control system for multi-agent runtimes, far beyond a simple terminal chatbot.
- Article Claude Code Source Code Walkthrough 01: From cli.tsx to main.tsx Dive into the Claude Code startup mechanism. We trace the system assembly from cli.tsx to main.tsx, exploring runtime environments and capability snapshots.
- Article Claude Code Highlights 05: Unified Access Control Claude Code centralizes decision-making via unified access control, balancing high-concurrency execution with robust user security and system oversight.
Agent Context Memory and Retrieval Guides and Resources
Study layered and external memory, agentic retrieval, RAG, and the effects of prompting on system behavior.
Remembering a preference, resuming a task, and retrieving a document are different context problems. Compare Claude Code and Gemini CLI memory designs, external memory, RAG, and agentic retrieval here. Personality and PUA prompting articles examine instruction effects; read their experimental limitations rather than treating a headline as a performance guarantee for every task.
Open this route’s guide →- Article Decoding Claude Code: 10 Highlights (Part 8) — The 3-Tier Agent Memory Scope An in-depth look at Claude Code's 3-tier memory architecture. By segmenting user, project, and local scopes, it prevents context pollution and boosts AI utility.
- Article AI Gets Gaslit: Boosting Performance by 50% with Corporate PUA Tactics Discover PUAClaw, an open-source project that turns corporate workplace manipulation tactics into effective prompt engineering for better AI performance.
- Article Google Just Flipped the RAG Table: Gemini File Search Deep Dive and Economics Breakdown Deep analysis of how Google's Gemini File Search simplifies complexity through managed RAG paradigm shift, plus economic model comparison with OpenAI.
- Article DeepSeek's New OCR Explained in Jianghu Terms: I Hereby Crown DeepSeek the Dark Arts Overlord DeepSeek-OCR revolutionizes traditional OCR with optical context compression technology, processing documents with extreme efficiency and showcasing a new AI input paradigm.
- Article Is RAG Dead? Claude Core Developer Proposes Agentic Retrieval Claude core developer proposes Agentic Retrieval, challenging traditional RAG paradigm — trusting AI's own reasoning ability is key to future AI engineering.
- Article Does AI Have MBTI? I Dissected Top Conference Papers to Create an AI 'Personality Injection' Playbook Deep analysis of top conference papers — master the practical handbook for injecting specific personalities into AI using MBTI, helping you build AI assistants with more soul and professionalism.
- Article Claude Code External Memory: The MemoryCog Architecture A Claude Code external memory architecture using MemoryCog, persistent Markdown logs, custom commands, and an eight-layer prompt for recording and retrieval.
- Article Deep Dive: Claude Code's 'Agentic Search' and Workflow Engine [Technical] A deep analysis of Claude Code's Agentic Search and Stateful Toolchain, revealing how it empowers efficient software development workflows by simulating human codebase exploration.
- Article Learning Agent Development with Google Gemini CLI (Part 3): How Does Gemini CLI 'Remember' Key Information? Learn how Gemini CLI manages global, project, and local memory through its hierarchical instructional context system for persistent AI cognition.
AI Coding Workflows Guides and Resources
Build a delivery workflow with Claude Code/Codex practice, Gemini CLI tools, frontend tests, and Vibe Coding for marketers.
AI coding begins with understandable requirements and ends with verifiable delivery. This collection links production Claude Code practice, Codex comparisons, Gemini CLI tooling, and Vibe Coding guides for marketers. Choose by task: start small tools with requirements and acceptance, examine tool boundaries for code changes, and validate frontend interactions under real viewport conditions rather than relying on generated screenshots.
Open this route’s guide →- Article Good LCP, Users Still Waiting: A Startup Performance Case Study from Our Real-Time Voice Web Tool Our real-time voice web tool had a 416 ms median LCP, yet the actual tool took 5.5 seconds to appear. Here is how we server-rendered the full workspace, moved service checks to the background, and cut tool appearance to 293 ms, with honest measurement caveats.
- Article Should Codex Users Switch to Hermes? My Deep Dive with GPT-5.4 Pro Don't rush to migrate from Codex to Hermes. Instead, treat them as a manager-worker duo, leveraging a hybrid architecture for better scheduling and coding.
- Article Browser AI Sidebars Are Eating Your Buttons: Lessons from Creem Browser AI sidebars often squeeze web layouts, hiding critical buttons. Learn why this happens and get four tips to optimize your responsive UI for users.
- Article Agentic Engineering Patterns: When Coding Becomes Cheap, Guardrails Are the New Ticket As AI lowers coding costs, engineering guardrails become the core competitive edge. Explore how Agentic Engineering Patterns integrate AI into workflows.
- Article Ladybird Ports LibJS to Rust in Two Weeks with Claude Code | Uncle Fruit's AI Daily Ladybird successfully ported LibJS to Rust in just two weeks using Claude Code and Codex, highlighting the power of AI-assisted engineering in code migration.
- Article Why 99% of Today's Hot AI App Generators Will End Up as Adult Electronic Toys Nine out of ten AI app generators will become toys because they can't solve commercial software's complexity, deployment gaps, and product logic challenges.
- Article Codex Deep Dive: Silent Assassin or High-Maintenance Waifu? A $200/Month User's Field Report Deep review of GPT-5.2 Pro and Codex, revealing Codex's stable execution versus Claude Code's fast interactivity, plus the Atlas+GPT workflow.
- Article GPT-5.2 Pro Frontend Test: Is 30-Minute Reasoning Worth $100+ Pricing? Hands-on testing GPT-5.2 Pro's frontend reproduction capabilities, revealing the staggering reasoning time and efficiency bottlenecks behind its premium price.
- Article Gemini 3.0 Pro Is Finally Here — Quick Test, UI Designers Beware? By Mr. Guo For vibe coders and indie hackers without design chops, the design handoff is often the biggest blocker—long before you worry about Gemini’s backend...
- Article Google Code Wiki Review: A Developer's Take on Grok and Code A hands-on Google Code Wiki review covering codebase documentation, Gemini context, diagrams, repository coverage, and its practical value alongside Grok.
- Article Gemini 3.0 Pro Is Here: A Quick Test That Should Worry UI Designers Gemini 3.0 Pro demonstrates impressive frontend CSS drawing and animation capabilities in stylized UI reproduction tests, signaling potential disruption to UI designers' workflows.
- Article Grok Is Busy Explaining the Universe, Google Code Wiki Helps You Understand Code First Google Code Wiki uses AI to automatically explain codebases, solving the developer pain point of reading 'legacy code' — value far exceeding Grok's grand narratives.
- Article Code as Music: A Musician's View of Software Engineering, Product Management, and AI Collaboration Musical aesthetics and software engineering share underlying logic — AI helps musicians transfer structural thinking to code, enabling efficient 'Vibe Coding.'
- Article Claude Code + Playwright MCP Extension Cured My Screenshot Shaky Hand Syndrome Using Claude Code with Playwright MCP extension to automate processing large volumes of infographic card screenshots — finally free from manual screenshot hassles and jitters.
- Article Vibe Coding for Marketers (Part 2): From Excel to Python — Become AI's Data Chef Learn Python and Pandas to efficiently consolidate messy data, providing high-quality 'fuel' for AI to achieve deep data analysis and attribution.
- Article Vibe Coding for Marketers (Part 1): Ditch Dev Queues, Build Your First Growth Tool Master Vibe Coding to escape dev queues. Use natural language to command AI and build your first growth automation tool with your own hands.
- Article Why Every Marketer and Ops Professional Should Learn Vibe Coding Vibe Coding empowers marketers and ops professionals to break free from resource dependency, solve real problems with lightweight code, and become full-stack growth talent.
- Article 4 AI Squad Tactics That Transformed Claude Code from 'Outsourced Junior' to 'Full-Stack Tech Partner' Master four Claude Code 'AI squad tactics' to upgrade AI from code outsourcing to a reliable tech partner through systematic collaboration.
- Article How to Use Claude Code in China: Three Paths — One Will Suit You How to reliably use Claude Code from China? This article provides three paths: stable proxy, self-managed environment, and open-source alternatives.
- Article How I Upgraded Claude Code from 'Toy' to 'Production-Grade' Development Workhorse [Best Practices] Master the systematic workflow for upgrading Claude Code from toy to productivity workhorse — build an external brain and precise planning to achieve high-quality, maintainable engineering code output.
- Article Learning Agent Development with Google Gemini CLI (Part 4): How Does Gemini CLI Scrape Data? Learn to use Gemini CLI's google_web_search and web_fetch tools, mastering how to safely and efficiently let AI Agents scrape and process real-time web data.
- Article Learning Agent Development with Google Gemini CLI (Part 2): How Does Gemini CLI Read and Modify Your Code? Explore how Gemini CLI safely and efficiently reads, understands, and modifies your code through layered perception and precise replacement tools.
- Article Learning Agent Development with Google Gemini CLI (Part 1): How Is Gemini CLI's Sandbox Implemented? Explore how Google Gemini CLI implements cross-platform sandboxing through Docker and native macOS technology, ensuring system security when AI Agents execute commands.
Agent Costs Safety and Operations Guides and Resources
Evaluate long-running agents through task metrics, tokens, SSD writes, hardware, API risks, and hosting costs.
Persistent agents face limits in hardware, bills, and incident recovery. Agent evaluation, token use, the Codex SSD investigation, PC configuration, API risks, and Vercel cost analysis provide concrete checks; the storage tool supports local inspection. Establish measurement methods and workload intensity before choosing reversible actions. One incident or heavy automation setup should not be generalized to every user.
Open this route’s guide →- Article Codex SSD Writes: My Local Logging Investigation My Codex SSD write investigation covers local SQLite logs, analytics settings, measured write rates, trigger mitigation, tradeoffs, and rollback.
- Article Why I Advise Against Using Unofficial APIs: Lessons from a Former Seller Drawing from my experience selling them, I strongly advise against using unofficial APIs. They pose severe security risks that can compromise your production.
- Article PCs in the AI Era: Built for Agents First In the AI era, PC selection has shifted. Prioritize high RAM and sustained throughput over local model performance to power efficient, concurrent AI agents.
- Article Anthropic: A Study in Hypocrisy An in-depth look at Anthropic's risk management and model competitiveness. We explore the user experience of Claude bans and its rivalry with ChatGPT in coding.
- Article Vercel for Indie Hackers: Why Your Free Tier Might Lead to a Surprise Bill Don't let Vercel costs catch you off guard. Learn how to optimize caching, limit API usage, and secure your serverless functions to avoid unexpected bills.
- Article AI Safety Audit: Can AI Accidentally Cause Harm? | Uncle Fruit's AI Daily Can AI trigger accidental harm in high-stakes real-world decisions? We explore shifting AI safety from model alignment to system auditability and accountability.
- Article Why AI Agents Use So Many Tokens: Claude Code and Codex Why AI agents use so many tokens: Claude Code and Codex examples, tool calls, growing context, and practical ways to assess cost and ROI.
- Article AI Agent Interview: How to Answer and Land Remote Agent Developer Offers Master the three-layer Agent evaluation framework: resource efficiency, task effectiveness, and system robustness. Use quantified scorecards to drive development decisions and land offers.