Deep Dives into AI Engineering

Anthropic removed 80% of Claude Code's system prompt for newer models with zero regression. This post breaks down the six new rules — from trusting model judgement to progressive disclosure — with practical before/after examples for each.

A supplementary comparison adding codebase-memory-mcp to the Graphify vs GitNexus vs CodeGraph showdown. 158 languages, a Cypher query engine, embedded semantic search, and 120x fewer tokens — all from a single zero-dependency binary.

SpaceXAI open-sourced 250K+ lines of Rust after a privacy scandal. This post breaks down Grok Build's architecture — actor model, leader daemon, kernel sandbox, parallel sub-agents — and compares each system head-to-head with Claude Code.

Hermes Agent and Claude Code share the same architecture — agentic loop, tool registry, markdown memory, MCP, hooks. But they split on one fundamental question: what is memory for? A deep comparison of memory-as-influence vs memory-as-product in AI agent design.

Agent skills supercharge your coding workflow — but they also open doors to prompt injection, data theft, and remote code execution. Two concrete attack walkthroughs and a practical defense checklist.

A head-to-head comparison of three code knowledge graph tools that give AI agents structural awareness of your codebase — with clear winners per category and a decision framework for choosing the right one.

AI agents don't lack knowledge — they lack discipline. This post explores agent-skills: 24 structured workflows that constrain AI agents to follow production-grade engineering practices, with a Tetris experiment showing the dramatic difference.

From prompt engineering to loop engineering — how developers are shifting from crafting individual prompts to designing autonomous systems that find work, run agents, verify results, and persist state without human input at each step.

Turn Claude Code from a one-shot code generator into an iterative optimizer. This post introduces alphaevolve-skill — an open-source skill that brings evolutionary optimization to your coding workflow.

A single LLM call generates plausible code — AlphaEvolve wraps LLMs in an evolutionary loop with automated evaluators to discover optimal code. Understanding this distinction unlocks a new mental model for AI-driven optimization.

An introduction to Graphify — a tool that maps your entire project (code, docs, images, videos) into a persistent knowledge graph you can query instead of grepping through files.

Public benchmark scores are a poor proxy for real-world model quality. This post covers data contamination, benchmark overfitting, and how to actually evaluate models for production.

A deep dive into the architecture of Claude Code — Anthropic's agentic coding tool that can read your files, run commands, and make edits across your entire project.