Development

Three Months with AI: From Experiment to Daily Workflow Over the past three months, I’ve been intentionally exploring how AI can fit into my software development workflow. Here’s what I learned. Building with AI: The Prototype Experiment My initial foray into AI tooling was building a prototype to understand the fundamentals. I spent about 6 hours creating ca-bhfuil, a simple experiment in combining multiple AI components: LangChain and LangGraph as the agent framework Ollama for local LLM hosting sqlite-vec as a basic vector database for RAG experimentation The result was a very simple prototype using hardcoded data to demonstrate storing commit messages in a vector database, prompting an LLM to use RAG to retrieve relevant data, and generating responses. You can see an example response in the commit message. ...

Using AI to Write OpenStack Nova Specifications: A Real-World Experiment. Background and Motivation After attending the 2025.2 PTG and seeing the extensive backlog of specs that need to be written, reviewed, and implemented, I decided to run an experiment: can AI meaningfully help with the OpenStack specification process? Not just the writing, but the actual design thinking and technical architecture work that goes into a good Nova spec. The motivation was simple - we have more good ideas than we have time to properly document them. Between my day job, core review responsibilities, and the general time constraints we all face, turning a concept into a well-structured, technically sound specification can take days or weeks. What if AI could help accelerate that process? ...