Posts
All the articles I've posted.
-
AI Model Comparison, Mid-2026 Edition: Two Months After blog080, the Model Layer Has Turned Over
blog080 was written in early March 2026. Two-plus months later, GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro have all shipped, and open-source flagships GLM-5.1/Qwen 3 Coder have closed the gap to within 5-15 points of closed models. This is the May update: what changed, and how to adjust your March picks.
-
AI Tooling Supply Chain Security Checklist: 8 Defense Principles Distilled from the Vercel and Nx Console Incidents
Neither the Vercel breach nor the Nx Console incident was a protocol vulnerability—both were credential governance failures. This post distills these two AI tooling supply chain attacks into 8 defense principles plus a 1-hour audit checklist, covering OAuth least privilege, secret tiering, managed device isolation, and IDE extension credential isolation—a security playbook indie developers and small teams can act on immediately.
-
Claude Code Multi-Agent Orchestration Plugins Compared 2026: Choosing Between Ruflo, Maestro, Claude Octopus, and Codex Peer Review
A head-to-head comparison of multi-agent orchestration plugins: Ruflo calls itself the "leading Claude orchestration platform" but underdelivers in execution, Maestro stays lightweight, Claude Octopus runs reviews across 8 models in parallel, and Codex Peer Review gates merges behind three sequential reviewers. From architecture to measured token costs — a decision framework for indie developers.
-
Claude Code Workflow Plugins Compared (2026): Superpowers, Shipyard, Ralph Loop, Maestro, or Karpathy CLAUDE.md?
The Claude Code ecosystem has splintered into 100+ plugins as of May. This post zooms in on the "workflow methodology" category—Superpowers, Shipyard, Ralph Loop, Maestro, and Karpathy CLAUDE.md. Design philosophy, context overhead, fit, and combination strategies, plus a decision tree for indie developers.
-
Astro 5 to 6, Fully Documented: Real Migration Data from a 48-Page Blog — the Official "2x Faster" Claim Doesn't Hold for Small Blogs
I upgraded my own blog (48 pages, Astro 5.16.6) to Astro 6.3.1 and recorded what actually changed, whether builds got faster, and what broke. Verdict: near-zero migration cost for a small blog, but the official "2x faster" claim doesn't hold at 48 pages — measured build times were essentially flat.
-
AI Agent Persistent Memory Architectures Compared: File-Based vs Vector Retrieval, Benchmarked with a blog-preflight Subagent
I hooked the same Subagent up to both Claude Code's built-in file-based memory and mem0's vector retrieval, then compared token cost, recall quality, and cross-session learning. The result: concrete thresholds for which approach fits which data scale, plus a look at procedural memory—the weakest but most promising direction.
-
Claude Code's Five-Layer Architecture Explained: How MCP, Skills, Agent, Subagents, and Agent Teams Work Together
Anthropic officially describes Claude Code as a five-layer architecture: MCP for connectivity, Skills for task knowledge, Agent as the main worker, Subagents for parallel isolation, and Agent Teams for coordination. This post breaks down each layer's role and collaboration patterns, with a real-world example from my blog's blog-preflight Skill showing three layers working together.
-
Frontman Deep Dive: What an AI Agent Can Do When It Sees Your Code from the Browser, Paired with Frontend Skills
Cursor and Claude Code both start from source code, but a frontend engineer's real work happens in the browser—the actual color on hover, the real DOM after SSR, the re-render triggered by the third useState. Frontman works in the opposite direction: from the browser back to the code. This post breaks down its architecture and combines it with Anthropic's frontend-design Skill and others into a complete frontend AI workflow.