Dries Buytaert on Why AI Workflows Must Balance Structure and Intelligence
Dries Buytaert explores how enterprise AI systems must strike a balance between flexible intelligence and reliable execution. Prompted by a recent Salesforce failure, he examines how deterministic workflows play a critical role in maintaining trust and consistency.
After Salesforce’s AI skipped steps in a customer survey flow, Dries highlights why adding more AI is not always the solution. Instead, he outlines two architectural models: “outside-in,” where automation platforms control the process and call AI only when needed; and “inside-out,” where AI agents handle flows and rely on deterministic tools for precision tasks. Both models, he argues, are valid but serve different needs.
Dries points to open source tools like Activepieces and n8n as examples of systems that can manage complex workflows while offering auditability and control. He stresses that enterprises value accountability in customer-facing and compliance-heavy operations, and this is where deterministic layers become essential. The future of AI, he suggests, will be built on structured orchestration where each layer, workflows, AI decisions, and execution tools, has a clear role.


