Koen Verstrepen

AI builder + leader

I build agentic AI systems and large-scale recommender engines, from first principles to production. I have strong conviction on combining leadership with being hands-on. This site is where I share the patterns, failures, and lessons from systems I'm shipping.

Looking to trade notes with peers working on similar problems.

Read my writing Explore my projects

Latest writing & projects

February 22, 2026writing

#ai#tooling#developer-experience

Setting Up Agentic Coding Assistants for Your Team

How we configured Claude Code at Willow with coding guidelines, slash commands, and integrations with GitHub, Slack, and Notion.

February 22, 2026project

#ai#llm#developer-tools#github#go#cli

Building Codestyle: Teaching Claude Code Your Team's Conventions

How I built a CLI tool that extracts implicit coding conventions from your team's GitHub PR history and generates a CLAUDE.md for Claude Code.

January 10, 2026writing

#llm#synthetic-data#deepseek#ai-training#reinforcement-learning

How LLMs get smarter from self-generated data

Discussion on whether LLMs get smarter or dumber when they try to learn from their own data, based on empirical evidence from building DeepSeek V3.2.

December 18, 2025project

#ai-agents#web-crawling#rag#llm#infrastructure

Making AI Agents Browse the Web Blazingly Fast

Benchmarking managed web crawling APIs for AI agent pipelines. Spider.cloud's smart waterfall approach delivers 2x more content at 5x lower cost.

November 30, 2025writing

#agentic-ai#gemini#codex#claude#nextjs

How coding agents have built, screwed-up and fixed koenverstrepen.com

Building a personal site from scratch with AI coding agents: what I learned comparing Gemini 3, GPT-5.1, and Claude across different platforms.