I Automate Everything (effective-programmer.com)

In an insightful reflection on automation and AI, a tech veteran shares their approach to integrating large language models (LLMs) into automated workflows while emphasizing a controlled and deterministic execution architecture. Drawing from experience in release engineering at Oracle, the author critiques the current trend of viewing LLMs as autonomous agents. They argue that while LLMs excel at tasks such as summarization and data extraction, they should not be the sole decision-makers in automated systems due to issues of trust, observability, and vendor lock-in. The author proposes a structured automation framework featuring personas, tasks, and plugins. This architecture delineates clear roles: personas define contextual use cases, tasks ensure deterministic execution without relying on LLMs, and plugins offer reusable capabilities that interconnect various tools. By keeping automation processes transparent and debuggable, the approach encourages reliability and adaptability in an ever-evolving AI landscape. Ultimately, the author advocates for treating LLMs as sophisticated assistants that enhance human decision-making rather than as independent agents, thus ensuring greater control over automation workflows.