hi I'm Leonardo, a solutions engineer, builder and movie lover. when not working at MotherDuck, I enjoy building products. currently I'm focusing on two domains:
- how to build generalizable / easily adaptable agentic loops for economically meaningful tasks. it's is getting clearer that difficulty of generically implementing agentic loops to substitute "white collar" tasks is what is preventing the mass adoption of AI. anthropic's tasks and openai's decision to push for a model with a harness that "is harder to interrupt" (i.e. codex) are the first glimpses towards the future foundation for these loops. this means building solutions that solve problems that MotherDuck requires (issues reproductions, customer support) while experimenting with artifacts creation, reproducibility of the solution, saving traces of the context and reasoning used.
- enhancing the DuckDB ecosystem (disclaimer: MotherDuck is a cloud data warhouse built on DuckDB). this mostly is buildings and keeping up-to-date connectors and tools that were abandoned or didn't exist.
all my work leverages ai. current models in use are:
- openai's gpt-codex-5.2 (high/xhigh) via the codex harness - for complex tasks, this is my preferred tool
- opus 4.5 via the claude code harness - main workhorse, great for speed and tool/api use
- gemini 3 - ui and consistency checks for plans
I'm currently based in Amsterdam (NL). I appreciate remote collaboration for side projects, though I believe full-time work needs to be done in-person. I keep track of the movies I watch on letterboxed! and help(ed) organize weekly projections of classics and gems across Amsterdam's many independent (and often hidden) "projection rooms". A list of of these movies is here, being updated thanks to OpenAI's structured outputs.