June 2025 Summaries

Procter & Gamble's Senior Automation Manager, Adonis Castillo Cordero, shares how his team modernizes legacy data systems and powers AI workloads across a global enterprise using Apache Airflow as the orchestration layer. The team uses Airflow to connect disparate systems, transform, enrich, and store data for supporting AI and analytics workloads, while ensuring reliable and scalable pipelines. Adonis outlines four best practices for teams looking to scale their Airflow usage: dependency mapping, anomaly detection, tiered data quality monitoring, and keeping business intelligence dashboards simple and focused. The team uses a "carbon layer" architecture and leverages technologies like Apache Spark and Apache Kafka, with Airflow serving as the orchestration backbone that enables faster innovation with AI and analytics while managing legacy system complexity.

The competitive advantage in AI-driven analytics and applications is shifting from raw ingredients like code, algorithms, and data to the orchestration workflows that turn them into coherent solutions. The winners of digital transformation are those who built compelling abstractions on top of primitives, not necessarily with the best databases or search engines. Now, we're witnessing the next evolution: agentic transformation, where software isn't just reading from a database or writing to an index but "thinking," deciding, and acting in tight feedback loops. The raw ingredients are getting cheaper and better, but what doesn't commoditize is the choreography that binds those pieces into dependable outcomes. The competitive frontier has shifted from What model are you using? to How well do your agents cooperate under load, with guard-rails and retries? And success depends as much on how you chain models, tools, and data together as on the LLMs themselves. Orchestration is becoming the universal primitive of AI applications, and companies that build reliable, user-friendly orchestration layers will capture disproportionate value. The race isn't just to own use cases before the market realizes the convergence; it's to build the orchestration infrastructure that enables building a product so good that switching costs become prohibitive. Building on code-first orchestration platforms like Apache Airflow wins because they offer infinite extensibility, ecosystem leverage, operational maturity, and future-proof architecture. The Airflow AI SDK standardizes LLM calls as first-class Airflow tasks, giving users portability, composability, observability by default, guardrails, community gravity, and the power to own their graph.