Luciana Reynaud Ferreira·Apr 12The contract is the productBuild for what survives the next model—not what compensates for the current one
Luciana Reynaud Ferreira·Apr 10Your LLM costs are real, your unit of measurement is wrongHow per-call metrics corrupt routing, budgeting, and governance decisions
InGoPenAIbyLuciana Reynaud Ferreira·Apr 7Turnpike: a typed contract for LLM runtime economicsI built Turnpike to make LLM cost attribution auditable, portable, and explicit across runtime boundaries
Luciana Reynaud Ferreira·Apr 5How an architectural decision cut LLM inference costs by 50×$210/month → $4.20/month: reliability first, then scheduling, then cost
Luciana Reynaud Ferreira·Apr 2Why distributed tracing doesn’t model LLM agent observabilityTraces were built for request-driven architectures. Agent orchestration breaks every assumption they depend on. I’ve been building…
Luciana Reynaud Ferreira·Mar 30O envelope que falta entre a intenção e a demissãoEntre a intenção de substituir e o valor capturado, o que falta é instrumentação.
Luciana Reynaud Ferreira·Mar 18LLM cost is a union type, not a scalarMost LLM cost dashboards are directionally helpful and decision-theoretically unsafe. One field fixes that.
Luciana Reynaud Ferreira·Mar 16The Pareto set of metrics for production LLMsLangfuse, OpenTelemetry, and what separates signal from instrumentation
Luciana Reynaud Ferreira·Mar 14The human–LLM contractWhy context is not memory and why real proficiency demands governance
Luciana Reynaud Ferreira·Mar 10O conjunto Pareto de métricas para LLMs em produçãoLangfuse, OpenTelemetry e o que separa observabilidade útil de teatro de dados
