Originally published on LinkedIn.

The LLM Automation Fallacy

You're asking your LLMs to automate existing workflows? That's not just using a Ferrari to deliver newspapers—it's parking a fleet of supercars in your garage and only driving one.

The real power isn't automating ONE flow—it's creating a THOUSAND flows and selecting the optimal path.

Why use a single LLM when you can deploy multiple models simultaneously, each approaching your problem differently, then select the best result? You hesitate between Claude and OpenAI, between Grok and DeepSeek? Why not use all?

Traditional automation: "Make this process faster / cheaper."

Agentic thinking: "Let multiple AI agents explore different solutions simultaneously, then choose the superior outcome."

Are you still considering using a fleet of supercars to deliver newspapers?