Blog

Shadow code appears when teams ship AI-generated output that nobody has fully reviewed, understood, or owned.

Every style guide, naming convention, and architectural principle exists for the same reason: humans have to read the code. That assumption is starting to erode.

When AI coding results disappoint, the problem is often not the idea of AI assistance. It is the stale stack underneath: outdated IDEs, outdated plugins, and outdated model choices.

The closer an AI model runs to your data, the more private and the more expensive. That tradeoff is not going away — but it can be managed.

The claim that AI-assisted development makes microservices obsolete is gaining traction. It is partially right. It is also missing the point.

When your development workflow depends on cloud-hosted LLMs, provider outages become your outages. The availability problem is no longer just an ops concern.

Naming is architecture. Inconsistent naming quietly compounds complexity across systems, teams, documentation, and automation.

Why treating GitHub Copilot configuration as a one-time setup leads to silent quality decay — and how agents, skills, and hygiene loops keep it alive.

How AI coding systems are reshaping software roles from implementation to orchestration, verification, and business-aligned system design.

Welcome to the blog! Here I share insights into AI development, software engineering and innovative technologies.

A technical look at the development of the Formageddon Chrome Extension and how AI is revolutionizing form filling.