AI as a tool
I use agentic AI tools every day for both work and personal projects. I have done so since they first appeared, but my usage intensified as models, especially open weights ones, became more competent during early 2026.
AI is a tool, not a replacement for thinking, domain and product understanding, or engineering skills.
Engineering is more than coding: it's about designing systems; managing and minimizing complexity; caring for users, team members, and stakeholders; being adaptable and ensuring long-term stability; and ultimately about taking responsibility for the work being done.
As a consequence, I believe in keeping a close eye on everything an agent produces. This includes reading, reviewing, and understanding everything an agent does on my behalf, and refining the output until it meets my architectural, engineering, and coding standards.
Tooling
I strongly prefer open source tooling - mainly Zed, Pi, and OpenCode - and open weights models. We have reached a point where open weights models are capable enough for everyday work. But tooling choices follow the project and client's requirements, not the other way around.
Background and talks
I specialized in AI for my MSc and I've kept following the field ever since. I really enjoy discussing the subject and learning more about practical uses together.
- Agentic AI Introduction: I gave an almost two-hour talk about agentic AI at the Biophysics Department of Eötvös Loránd University, covering the model and tooling landscape, costs, privacy, and learning, followed by a hands-on practice session.
If you struggle to get started with AI or to get value out of it, let's talk.