Pablo Albaladejo

Writing

Notes on making AI work in production: LLM evals, observability, agentic development, and GEO. Every claim links to an artifact you can check.

A discovery file is a promise: level 5 without a fake card
A gate is only as real as its run: three of mine that were not
AI-DLC: the only gates that count are the ones code enforces
Fitting counterparty models from public APIs
A 90% pass rate is not a gate: four criteria, two thresholds
A tolerant OAuth server validates your broken client
Evaluating LLM outputs in production: 22 benchmarks, a 90% gate
GEO: how AI agents find you, a same-day case study
Observability for LLM pipelines: from streams to sinks
Shipping production software with AI agents

ALSO PUBLISHED

Writing on other platforms.

Stop Chatting, Start Specifying: Spec-Driven Design with Kiro IDE ↗ Kiro blog
Observable AI Streaming on AWS ↗ dev.to · 4-part series