The original title is "AI Is Great at Reasoning. Stop Using It for Workflows."

Short summary

A platform engineering team shares lessons from building and then removing AI-generated Terraform code at runtime. The key insight: LLMs excel at reasoning through ambiguous inputs at system boundaries, but deterministic pipelines should remain deterministic once intent is structured. Introducing probabilistic models into predictable workflows adds unnecessary risk.

• •LLMs add probability; deterministic automation stays 100% predictable
• •Use AI at system boundaries for natural-language-to-structured-data conversion
• •Don't let a probabilistic model decide which tools to run in production

Generated with AI, which can make mistakes.