AI & Agentic Development

I use Claude Code and OpenCode (a Claude CLI) as my main development tools. Not as a "copilot" that autocompletes lines, but as an agentic system that can reason about code, make changes across files, run tests, and iterate on its own work. The difference is huge. Instead of me typing code and an AI suggesting the next line, I describe what I want and the agent figures out how to get there.

At Pyango, we've pushed AI adoption across the whole team. Our best engineer, Melek, runs what I'd call an AI-first workflow. He uses Claude for code generation, testing, documentation, and review. The output quality and speed are noticeably different from team members who haven't fully adopted these tools yet.

I also run an internal workshop series to get the rest of the team up to speed. The plan is to record these sessions, turn them into content, and eventually offer them externally. Every agency and SME in Switzerland has the same problem: their people are too slow with AI tools because nobody taught them properly.

One of the more interesting things I've built is a personal knowledge management system in Obsidian that uses LLMs as the capture and connection layer. I call it the FLAP workflow: Fleeting notes and Literature get processed into Atomic Notes, which feed into Projects.

When I capture something (an article, a podcast, an idea from a conversation), Claude ingests it, writes a summary in my voice, creates wiki-links to related notes, and generates spaced-repetition flashcards. It also explains why things connect, which is where the real value is. I've found connections between trading strategies and neuroscience, between agency operations and systems thinking, that I wouldn't have spotted on my own.

The long-term goal is a densely interconnected knowledge graph that surfaces unexpected relationships across everything I learn and work on. It's getting there.