Agentic AI: Why You Need to Know Your Process First

What do AI agents need for AI to really work in your company? Most of the time, it fails because of data, missing interfaces or a process that nobody ever defined properly.

In this episode of it depends, we look at what an AI agent in your company needs: good data and capabilities that are reachable through interfaces. I explain why a composable architecture based on the MACH principles is the shortest way there, and why the agent then simply becomes the next channel alongside shop, app and POS. Plus a story from a client project with an SAP Commerce monolith, where exactly this idea turned the mood in the workshop around. And we clarify whether you need MCP or whether your APIs are enough.

Then we get to the part where I see most projects stumble: process analysis. You will learn our four steps (observe, break down, calculate, collect the truth), how to use five criteria to find the right process for your first agent, why you need a golden dataset and how to turn up autonomy step by step, from shadow mode to full automation. I also share how we at foobar started with agentic coding ourselves and what we learned the hard way.

At the end, you have four questions to ask yourself before you think about the right model or the right agent tool.

Missed part 1? What an AI agent actually is and what it is made of: What Is Really Going On With AI Agents?

Want to know which process is the right first candidate for an AI agent in your company? In the Agentic Impact Check by foobar Agency, we run the process analysis with you in two weeks, evaluate the candidates and test real cases on a prototype. At the end, you have the blueprint for your first agent: https://foobar.agency/en/impact-check

#AIAgents #AgenticAI #ComposableCommerce