I'm Bogdan, a software engineer in Bucharest. I currently work at Google on the Android windowing tools team, where I build tracing infrastructure with Perfetto for the graphics and windowing stack, from SurfaceFlinger to WindowManager.
Before Google I spent ~4 years at DataCamp on the Content Platform team, and before that Bloomberg in London on the MARS risk platform.
I post scribbles, short timestamped thoughts, and essays, the longer pieces.
One of the biggest wins of coding agents is that I can now be extremely thorough with experiments without me having to do all of the manual experiments. For example, I recently had to optimize a proof of concept I had built at work. I had 5 ideas, let's denote them optimization A through E. They were all pretty independent of each other, but each were more and more invasive and risky. So I wanted to get the most performance with the least amount of code health impact. Cue the agents. I had a swarm build all of the optimizations, and run experiments to test for jank, regressions, etc. Each test run takes about 3 hours of lab time, with constant monitoring. The agents handled most of the work that would have been pretty daunting for me to do beforehand. I also didn't stop there. I had it run combinations of all the optimization, e.g. A with D; B, C, and E; etc. Took about 3 and a half days of lab time, but I ended up with a great solution. I am sharing this thought because I believe this is one of the great use cases of agents, making them drive experiments that would have been extremely tedious for us to run in the past.