Claude Code Harness Audit: Finding Silent Failures in Hooks, npx, and YAML
A three-month drift audit of a Claude Code harness. Learn which failures stay silent and how to test hooks, avoid npx squatting, and validate YAML.
Blog
Hands-on notes on AI agents and work automation: guides, concepts and what actually worked.
- Insights
- Insights
When Can AI Agents Run Autonomously? A Verifier and Context Test
Autonomous agent loops need a machine verifier and explicit context. Learn the two-axis test, the evidence, and when humans should write rubrics first.
- Insights
LLM Wiki Method vs Google OKF: What Is Shared and What Differs
Compare a markdown-based LLM wiki pipeline with Google's Open Knowledge Format v0.1. Learn where they match, where they differ, and when to align your schema.
- Insights
Why Domain Expertise Matters More Than Coding Skill With AI Coding Agents
Anthropic's analysis of about 400,000 Claude Code sessions shows domain knowledge amplifies agentic coding results more than coding training does.
- Insights
MCP vs SDK: Why Protocols and Building Frameworks Are Different Layers
Why MCP and SDKs are complementary layers, not rivals, and how Google and n8n each handled protocol support versus building a framework.
- Insights
Air-Gapped Automation Delivery Playbook: Lessons from Traditional SI Methods
A four-gate delivery playbook for automation projects in closed networks. Catch tool limits and security constraints in week one, not week three.
- Insights
Fable 5 vs Opus 4.8: Real Session Cost and Agent Behavior Compared
Fable 5 costs twice as much per token as Opus 4.8, but measured sessions cost 1.0 to 1.4 times as much. See the token, cost, and intervention data behind it.
- Insights
Claude Code Hooks: Why Deterministic Control Points Work for Stochastic Agents
Why Claude Code hooks guarantee deterministic control over AI agents, traced from git hooks and Design by Contract to neuro-symbolic guardrails.
- Insights
MCP Ecosystem in 2026: How the Protocol Moved to Shared Industry Infrastructure
Why MCP moved to Linux Foundation governance, how it complements Google's A2A, and what the adoption and security evidence supports.
- Insights
Which IT Term Fits a Paused Service: Teardown vs. Mothballing, Turndown, and Decommission
Teardown means destroying one-off resources, not keeping them for later. Here is which term fits a service you shut down temporarily and plan to restart.