About
Carlos Tapia
Software architect. Thirteen years on systems for banking, telecommunications and the public sector. I have worked with AI since ChatGPT came out at the end of 2022, and with agents since 2026.
I work on critical systems in banking, telecommunications and the public sector: event-driven architecture, microservices, and integration with legacy platforms that can't be switched off while you replace them.
At work I use AI agents every day. None of that is on this site and none of the measurements come from it. The measurements come from agent teams I ran on my own time and budget (eight by September 2026), six of them on my own production systems: a point-of-sale product, a loyalty platform, a public website and others.
I started logging every turn and pricing it because I wanted to know which advice about running agents held up on my own systems.
The systems I measure are small, small enough to instrument completely. That is what makes the measurements possible, and it is also their limit: I don't know whether the numbers hold at ten times the scale.
Get in touch
If you run agents against real code and measure it, I'd like to compare numbers. If you found an error in a note, tell me.
hola@carlostapia.cl