Blog

This blog covers practical AI notes, engineering culture/decision-making, and occasional thought pieces.

• Follow the newest posts here
• Browse by theme → Tags

How to browse

Each post has at least two tags: pillar-* (topic) and type-* (format).

• Topics: pillar-ai / pillar-engineering / pillar-mkdocs
• Formats: type-howto / type-implementation / type-review / type-news / type-opinion / type-reference

Latest Articles

Weekly / Monthly digests

August 2026

• Before Enabling Codex's 1M Context for GPT-5.6 Sol: Official Settings vs. an Observed Cap - 2026.08.23
• Codex Remote Can Resume Existing Threads but Cannot Start a New Chat: 'Can't Verify Project Trust - 2026.08.22
• Why Codex Weekly Usage Can Increase Less Than Expected: Caching Is Not the Whole Explanation - 2026.08.22
• GitHub Outages in 2026: A Resilience Plan and the State of Forge Alternatives - 2026.08.20
• Was SpaceX's Cursor Acquisition a Side Note? How Origin Could Loosen GitHub's Grip - 2026.08.20
• Claude Code vs Codex: Which Is Better in August 2026? - 2026.08.16
• Choosing GitHub Copilot Models in 2026: Routing Eight Models by Effective Cost - 2026.08.14
• What Claude's AI Watermark Means: Article 50, Machine-Readable Marks, and Their Limits - 2026.08.11
• Inside Cloudflare OS: What to Decide Before Giving Every Employee an AI Agent - 2026.08.10
• Can Seedance 2.5 Really Deliver 4K and 180 Seconds? Reading the Official Spec Gaps - 2026.08.09
• Claude Code Cross-Session Messaging: Permission Boundaries and Organization Controls - 2026.08.09
• Why AI Token Budgets Run Out: Resource Allocation in the Credit Era - 2026.08.09
• What Was ChatGPT's Chat / Work Split? Current Modes and the Year-End Integration Comment - 2026.08.09
• Meta Muse Code Benchmarks: Did Muse Spark 1.2 Catch Claude Opus 5 and GPT-5.6? - 2026.08.06
• Why Giving Every Employee AI Breaks Down: From License Management to AI FinOps - 2026.08.05
• What Codex's Weekend /fast Pattern Means for Agent Mode Selection - 2026.08.03
• Instruction Volume Alone Cannot Measure LLM Performance: Separate Freedom from Measurement - 2026.08.01

July 2026

• Did Claude Mythos Change Cryptanalysis? HAWK's Withdrawal and the Reduced-Round AES Result - 2026.07.31
• Do not make Codex Security CLI a required check on day one: From first scan to CI - 2026.07.31
• Claude Accessed 3 Organizations: Anthropic's Isolation Failure - 2026.07.31
• Did Microsoft Really Beat Mythos? Checking MAI-Cyber-1-Flash's 95.95% CyberGym Score - 2026.07.29
• Claude Share Links Appeared in Google Search: How to Audit and Revoke Access - 2026.07.28
• Claude Opus 5 Benchmarks: Reading the Numbers Against Fable 5 and GPT-5.6 Sol - 2026.07.25
• ChatGPT Voice Comes to Desktop for Multi-Agent Orchestration - 2026.07.25
• Why Codex Weekly Usage Drains So Fast: Separating the June Incident from July Limit Changes - 2026.07.24
• Claude Security Comes to Claude Code: Multi-Agent Vulnerability Scans and Reviewed Patch Suggestions - 2026.07.23
• How an OpenAI Model Evaluation Reached Hugging Face Production - 2026.07.23
• What It Takes to Run Hugging Face speech-to-speech Fully Locally - 2026.07.23
• Does Gemini 3.6 Flash Really Cut Costs? The 17% Token Claim and Three Model Roles - 2026.07.22
• What Is Graph Engineering? How It Differs from Loop Engineering and Whether the 'Obituary' Is True - 2026.07.20
• Where Is Your AI Adoption Stuck? Turning One Person's 10x Into Organizational Results - 2026.07.19
• What Is MarkPDFDown? Converting PDF Pages to Markdown with a Vision LLM - 2026.07.19
• 1Password for Claude: Zero-Exposure Login for AI Agents - 2026.07.18
• Kimi K3: Where the 2.8T “Open 3T-Class” Model Beats—and Loses to—Fable 5 - 2026.07.17
• Where should AI security review stop? A three-stage design for work-in-progress, pull requests, and remediation - 2026.07.17
• What Is LiteLLM? One Gateway for 100+ LLM APIs—and the “Key Ring” Risk Exposed in 2026 - 2026.07.17
• What Is herdr? The 'tmux for Agents' That Runs Multiple Coding Agents in One Terminal - 2026.07.16
• GPT-Red Explained: OpenAI's Self-Play Attacker and GPT-5.6's Stronger Defenses - 2026.07.16
• Codex CLI Voice Input Was Removed in 0.118.0: Use Desktop Dictation Instead - 2026.07.15
• Why Codex on Windows gets slow with WSL: measured EISDIR retries and MCP growth - 2026.07.14
• Where to Put Stop Conditions in Codex Automation - 2026.07.12
• Exception-handling ownership to define before scaling AI adoption company-wide - 2026.07.11
• GPT-5.6 Sol vs Claude Fable 5: Why the Scores Split and How to Route Real Work - 2026.07.11
• Same Word, Different Mechanism: Claude Code vs Codex Context Compaction - 2026.07.10
• GPT-Live: ChatGPT Voice Stops Being Turn-Based - 2026.07.09
• Claude Code /checkup: Let AI Diagnose and Clean Up Configuration Debt - 2026.07.09
• Money Forward's GitHub Breach, Explained: What the 60,000 Records Mean - 2026.07.08
• There Are Four Levels of Delegating to AI: Reading Claude Code's Official Loop Guide Through a Familiar Example - 2026.07.08
• What Are Claude Code Dynamic Workflows? Moving Plans into Code and Running Subagents in Parallel - 2026.07.01
• Claude Sonnet 5 vs Opus 4.8: Price, Limits, and Benchmarks - 2026.07.01

June 2026

• WSL Containers Preview: WSLc vs Docker Engine and Compose Limits - 2026.06.30
• GitHub Actions Step Parallelism: When to Use It Instead of Splitting Jobs - 2026.06.29
• Publishing a Website for Free: How to Choose GitHub Pages or Cloudflare - 2026.06.28
• GPT-5.6 Preview: Sol / Terra / Luna, new modes, pricing, and government-reviewed rollout - 2026.06.28
• How to debug a missing Codex Automation PR - 2026.06.27
• Information to separate before mixing long context and RAG - 2026.06.26
• The Loop Becomes the Unit of Work: Long-Running Task Design in OpenAI's Codex White Paper - 2026.06.24
• Safety settings to create before running Codex for long sessions - 2026.06.24
• What Is OfficeCLI? How AI Agents Create and Validate Office Files - 2026.06.20
• Open Knowledge Format (OKF) Reviewed: What Is New in Google's AI Agent Knowledge Format, and What Is Verified? - 2026.06.16
• Fix Codex App database is locked in WSL mode: cause and workarounds - 2026.06.13
• Claude Fable 5 and Mythos 5 Paused for All Users: What to Check Now - 2026.06.13
• Claude Fable 5 Launch: What General Availability of a Mythos-Class Model Means - 2026.06.10
• Before expanding Codex beyond engineers, decide who owns the output - 2026.06.03
• Rethinking Harness Engineering through Core and Shell - 2026.06.03

May 2026

• Distributing Agent Skills: One Standard, Many Delivery Layers - 2026.05.31
• Audit Log Design Before AI Adoption Increases Operating Cost - 2026.05.31
• Why AI Adoption Stops at PoC: Separate Specification from Operations - 2026.05.30
• Codex Now Supports Remote Control on Windows: Your Phone Becomes a Console for a Windows Dev Machine - 2026.05.30
• Fix fragile Excel PDF and PowerPoint files before giving them to AI - 2026.05.30
• Separate AI Decisions From Human Decisions in Enterprise AI - 2026.05.29
• How to start Codex automation where humans only review PRs - 2026.05.26
• RAG Quality Depends on Ranking: Rerankers as a Way to Reread Search Results - 2026.05.20
• GitHub's 3,800 Internal Repository Breach: How a VS Code Extension Became the Entry Point - 2026.05.20
• Claude Code at Scale Vol. 1: Redirect to Japanese Article - 2026.05.17
• Amazon Bedrock Architecture for Enterprise: Where Models Run, Where Data Flows, and How Custom Models Fit - 2026.05.15
• Codex Web Response Latency Has Worsened in Measurable Public Signals Since May 2026 - 2026.05.13
• Enterprise AI Belongs in the Decision Loop, Not Just on a Dashboard - 2026.05.11
• Measuring AI Review Value with Copilot Code Review Metrics: KPI Design by security and bug_risk - 2026.05.09
• Are Pull Requests Outdated? Three Change-Management Options for the AI Agent Era - 2026.05.08
• What Is Codex /goal? How It Replaces \"Keep Going\" Prompts - 2026.05.05
• The Current State of Harness Engineering Terminology — A Map as of April 2026 - 2026.05.01

April 2026

• The AI-Surfaced Linux Kernel Vulnerability “Copy Fail”: CVE-2026-31431 Impact and Response Guide - 2026.04.30
• LLM Wiki Architecture: Karpathy's Markdown Knowledge Base Pattern Explained - 2026.04.29
• A Beginner's Guide to Three RAG Architectures: Classic RAG, Graph RAG, and Agentic RAG - 2026.04.26
• How NTT, Fujitsu, and NEC Signal Different Enterprise AI Strategies - 2026.04.26
• Recursive Language Models Explained: Moving Long Context into an External Environment - 2026.04.24
• JSX Article Publishing PoC: Put a React Visual Article Under Blog - 2026.04.24
• Claude Code's Quality Drop Was Not Imaginary: 3 Causes from Anthropic's Official Postmortem - 2026.04.24
• gh-stack Explained: Why Branching Strategy Changes in the Age of AI Multi-Agent Development - 2026.04.18
• What Is Claude Mythos? Why Anthropic Won't Release It Publicly - 2026.04.17
• Which Work Should AI Actually Handle? A Practical Guide to Business Application Design - 2026.04.14
• Where Does Value Move in the AI Era? From Using AI to Making AI Work - 2026.04.13
• Why Claude Code Burns Through Tokens So Fast — 3 Causes and the Cache Bug Confirmed by a Source Code Leak - 2026.04.01

March 2026

• OpenAI Releases Official Claude Code Plugin — What codex-plugin-cc Means - 2026.03.31
• Codex vs Claude Code: The Divergence in Subagent Design Philosophy - 2026.03.28
• What Is Harness Engineering—What Do You Actually Build? - 2026.03.27
• Azure Skills Plugin: A New Architecture That Injects 'Azure Judgment' into AI Coding Agents - 2026.03.27
• Did CLI Really Beat MCP? Deconstructing the Debate's Premises - 2026.03.23
• Skill Obsolescence Is Inevitable: Detection, Recovery, and Localization in Operations Design - 2026.03.20
• Should You Take Claude Certified Architect? Scope, Price, and Validity - 2026.03.17
• Don't Keep Specs—Connect Them - 2026.03.15
• Lessons from Amazon's Outage: What Organizations Moving Fast with AI Were Missing - 2026.03.12
• Why AI Coding Breaks Team Development — The Explosion of Local Optima and Machine-Readable Consensus - 2026.03.10
• Will Engineers Stop Growing in the AI Agent Era? — Research Data Reveals the Structure of Skill Atrophy and Talent Pipeline Collapse - 2026.03.05
• The Truth Behind +11% Software Engineer Job Postings — Not a Recovery, but a Transformation - 2026.03.04
• Automating the Claude Code × Codex Review Loop — Three Levels: SKILL.md, Plugin, and Pipeline - 2026.03.02

February 2026

• Trump Bans Anthropic from Federal Use — The Dispute and the Reality of OpenAI's Deal - 2026.02.28
• What Benchmarks Reveal About Cross-Model AI Coding: Why Combining Claude Code and Codex Beats Going All-In - 2026.02.28
• Breaking: Anthropic Launches Claude Code Security — AI Detects Code Vulnerabilities Like a Human Researcher - 2026.02.21
• When AI Can Teach You How to Use AI, What Are AI Courses Really Selling? - 2026.02.20
• Context Engineering: Comparing RAG, MCP, and Agent Skills — A 2026 Design Guide - 2026.02.19
• Claude Sonnet 4.6 Release — Opus-Level Performance at Nearly Half the Cost - 2026.02.18
• What Is OpenClaw? A Sober Look at the Viral Autonomous AI Agent - 2026.02.17
• Structuring the 'SaaS is Dead' Debate: What Dies and What Survives in the AI Agent Era - 2026.02.08
• GPT-5.3-Codex Complete Guide | Terminal-Bench 77.3%, 25% Faster, Codex App Usage & Adoption Decisions - 2026.02.06
• Claude Opus 4.6 Complete Guide (Feb 2026) | Pricing, 1M Context, Benchmarks & Usage - 2026.02.06
• How to Fix an Agent Skill That Does Not Trigger: Description Design and Tests - 2026.02.02

January 2026

• Is the Age of MCP Over? Tool Integration Design in the Age of Skills Compilation - 2026.01.27
• SkillsMP Review 2026: What It Is, 1.6M+ SKILL.md Files, and How to Choose Safely - 2026.01.18
• Spec-Driven Development in the AI Era: Stop Stalling with Output Contracts - 2026.01.10
• Claude Code Vulnerability CVE-2025-64755 | Sed Command Validation Bypass Enables Arbitrary File Writes - 2026.01.09
• Claude Code Agent Swarms Complete Guide | 3x Development Efficiency with MCP Integration - 2026.01.03
• Claude Agent SDK Beginner Guide: Build a Read-Only Python Agent - 2026.01.01

December 2025

• Agent Skills Installation Guide 2025: 10 Practical Examples + Safety Checklist - 2025.12.31
• Agent Skills in Practice: where it works, benefits, and how to measure - 2025.12.23
• Agent Skills Quickstart: pre-built vs SKILL.md in 3 minutes - 2025.12.23
• Agent Skills: Why SKILL.md Won't Load + Fix Guide - 2025.12.21
• The MCP vs Agent Skills Debate: Untangling the Category Error - 2025.12.17
• Breaking: GPT-5.2 Release - 400K Token Context & 30% Hallucination Reduction - 2025.12.11

November 2025

• Scrum × AI Coding: A Practical Guide to Spec Management in Agile Environments - 2025.11.29
• Should You Write Specs? 3 Patterns for Specification Management in AI Code Generation Era - 2025.11.28
• Claude Opus 4.5 Complete Guide - 76% Token Efficiency Improvement Balances Peak Accuracy and Cost Reduction - 2025.11.25
• Gemini 3 Agent Development Guide: Implementing thinking_level and Autonomous Execution - 2025.11.24
• Nano Banana Pro Complete Guide: 10 Use Cases for Google's Latest AI Image Generation [2025 Edition] - 2025.11.23
• Gemini 3 Pro In-Depth Review: Performance and Real-World Usage Analysis - 2025.11.18
• MCP Code Execution Deep Dive: Agent Design Achieving Up to 98.7% Token Cost Reduction with Code Execution with MCP - 2025.11.16
• GPT-5.1 Complete Guide: Evolution from GPT-5 and Practical Selection Strategy - 2025.11.13
• Kimi K2 Thinking: Chinese Open-Source AI Surpasses GPT-5 in Key Benchmarks - 2025.11.13
• Japan AI Basic Plan Complete Guide - 2025.11.13
• Claude Haiku 4.5 Practical Guide - Achieve Sonnet 4-Level Performance at ⅓ Cost - 2025.11.13
• OpenAI Sora Android Release - The Official Launch of a Text-to-Video AI App - 2025.11.08
• OpenAI Codex CLI Updates (November 2025) Summary - 2025.11.01

October 2025

• AI Bubble Market Strategy Dashboard - 2025.10.17
• The Deep Layers of the Readable Code Debate: Human Anxiety and Values Behind Technical Arguments [2025 Edition] - 2025.10.05
• AI Era 'Readable Code Obsolescence' Practical Decision Guide - 2025.10.05
• The Full Picture of OpenAI Sora 2 Copyright Issues - Essential Countermeasures for Japanese Creators - 2025.10.03