Claude Code, Cursor, Copilot: Architecture, Not Autocomplete
AI coding assistants stopped being autocomplete once they could read a codebase and plan multi-file changes. Here is what actually changed in how I work.
Writing about frontend systems, automation, AI workflows, and things I figure out while building.
AI coding assistants stopped being autocomplete once they could read a codebase and plan multi-file changes. Here is what actually changed in how I work.
Local LLMs via Ollama trade raw capability for privacy, cost, and offline access. Here is how I actually decide between local and API-based models.
A practical filter for deciding what belongs in an n8n + Groq workflow versus what stays manual, based on the automations I have actually kept running.
AI writes implementations faster, but that raises the value of tests instead of lowering it. Here is the testing discipline I apply to AI-generated changes.
Vibe coding gets code running fast, but running and correct are different claims. Here's the gap I actually check before shipping AI-generated code.
What cubic-bezier() actually controls, why the built-in easing keywords look different from each other, and how to read or build your own custom curve.
How I decide between simple and scalable architecture early in a project — file-based content, CMS-ready layers, and when to actually add complexity.
Reusable n8n + Groq automation patterns — prompt chaining, structured output, and rate-limit handling — from workflows I actually run.
Practical screen-reader accessibility checks for React apps — semantic HTML, focus management, ARIA landmarks, and how I actually test them.
Category