I write about the parts that break after the demo.
Most of what I care about isn't the raw model capability. It's everything that surrounds the model in production: the run that costs $0.04 more than the last one for no discernible reason, the tool loop that recursed silently until the budget was drained, and the agent shipped without a kill switch.
"If your agent system cannot produce a versioned, unit-level cost ledger and an instant circuit breaker for every execution step, you do not have an autonomous system. You have an unmetered bill."
Originally from Portugal, currently living in Amsterdam, learning Dutch. Spent the last 8 years shipping distributed software and backend infrastructure across Europe.
Operator-grade reliability, unit-cost accounting & release-gating for AI agents.