> adrianweb

Blog

Thoughts on Java, cloud architecture, and building software.

  • August 15, 2026

    Building Focusly: A Working Demo of AI Atomic Feature Architecture

    Focusly is a working todo application with an AI assistant through MCP or local Ollama—and a practical proof of AI Atomic Feature Architecture.

  • August 5, 2026

    AI Atomic Feature Architecture

    What if software projects were designed so an AI agent could implement one small feature without understanding the whole codebase? That's the idea behind AI Atomic Feature Architecture.

  • April 17, 2026

    AI is Probabilistic. Your Business Logic Shouldn't Be.

    AI is incredible, but it doesn't calculate. It guesses. When your business depends on exact numbers and reliable processes, guessing isn't good enough. Here is what AI is actually great at and where it will silently fail you.

  • April 8, 2026

    AI Generates 50 Tests. You Need 5.

    AI coding agents write too many tests, most of them bad. They mock everything, test implementation instead of behavior, and break on every refactor. There's a better way — and it starts with treating tests as specifications.

  • March 21, 2026

    Open-Source Agentic AI Frameworks — The Comprehensive Guide (2026)

    22 frameworks compared — CrewAI, AutoGen, LangGraph, Mastra, Vercel AI SDK, and more. Pros, cons, local AI support, and a decision guide to help you pick the right one for your project.

  • February 15, 2026

    Piwi and the Agentic Economy

    The next wave of AI isn't chatbots — it's agents. Here's how Piwi.ai is building with MCP, A2A, and open standards to become a native participant in the $6 trillion agentic economy.

  • February 15, 2026

    Open-Sourcing DocSchema — Why I'm Giving Away the Knowledge Layer

    DocSchema is an open standard that defines what data lives inside business documents. Here's why I open-sourced it, and how agentic AI coding made it possible.

  • February 14, 2026

    Building Piwi.ai

    Piwi.ai didn't start as a developer side project. It started from watching real people struggle with copy-pasting data from documents and the pain of verifying everything is correct.