Programming
The New Math of AI-Native Engineering
Why fast agents still ship slowly
Studio room
11 November 2026, 10:30–11:00
AI-native workflow guides tell a familiar story: code is no longer the bottleneck, so redraw the process as a loop and put humans above it. What they rarely give you is a number. This talk takes the loop at its word and does the arithmetic. What happens when agents multiply the number of changes but the people who approve them stay the same? How much does a second AI reviewer really add? How many test cases do you need to notice your agents got worse? Four numbers, a few simple models you can rerun with your own data, and a Monday-morning checklist for measuring them in your own teams.
Alexandru Diniș
BMW TechWorks Romania
I started my tech journey 14 years ago as a Java developer and learned the stack the hard way: keeping up with Java versions, adapting to new frameworks, and moving from solid monoliths to modern, distributed systems.
Most of my work has been in the automotive field, where reliability matters and the smallest details add up.
Over time I moved into solution architecture and delivery management, making sure ideas turn into real, shipped software — and from there into technology leadership.
Today, as CTO, I'm handling the technical direction over 800 engineers: the architecture we bet on, the platforms we build on, and the engineering standards we hold. I'm also responsible about people: growing engineers, aligning with stakeholders, and making sure what we build actually moves the business.
I'm still hands-on. I care about Java, Cloud and AI, and about finding real ways to integrate them into the development lifecycle, from design to delivery. DevOps practices and security are part of the same conversation, not an afterthought. Beyond that, I enjoy exploring, writing about what I learn, and sharing practical notes and lessons from the field.
