system_log .archive

A collection of technical findings, architectural patterns, and notes from building things that actually run.

Status: Indexing_CompletePage: 01/02Entries: 004/008

ID: 0012026.03.30

LLMs are finding zero-days now. The 20-year balance just broke.

2026.03.30

Nicholas Carlini showed what I didn't want to believe: Claude Code, running with no fancy scaffolding, is autonomously finding heap buffer overflows in the Linux kernel that sat undetected since 2003. The attacker-defender equilibrium we relied on for two decades is over.

#security #ai #llms #vulnerability_research #zero_day

ID: 0022026.03.24

Kiro vs Claude Code (2026): Spec-Driven vs Terminal-Native AI Coding

2026.03.24

Kiro makes you write a spec before touching any code. Claude Code starts executing the moment you ask. Here's what each gets right, what each is missing, and when to use which.

#ai #tooling #developer_tools #claude_code #kiro #comparison #agentic_ai

ID: 0032026.03.22

When Claude Code stopped feeling like autocomplete

2026.03.22

Superpowers was the thing that changed how I use Claude Code — not because it's faster, but because it forces a structurally different workflow.

#claude_code #agentic_coding #developer_tools #productivity

ID: 0042026.02.03

The Road to 10x: 1. Debunking the Myth

2026.02.03

The '10x developer' label survives because it gets the output metric right and everything else wrong. Here's what high-impact development actually looks like.

#development #career #10x_journey