What Is an AI Agent, Really? Workflow, Prompt or Agent Washing

Everyone is talking about AI agents, but hardly anyone means the same thing. Gartner estimates that of thousands of agentic AI vendors, only about 130 deliver real agentic capabilities. Gartner calls the rest agent washing. Matthias Dietrich and Philipp Krueger clarify what an agent actually is.

Between a fixed workflow with a few LLM steps and a large prompt that launches its own subagents lies an entire autonomy scale. Using order intake as an example, Matthias shows where an agent may decide on its own and where the process has to stay rigid. Philipp explains why bad data usually comes from bad processes.

How much repeatability does your process need? Who is responsible when an agent runs on its own in the data center? And when is a simple IF the better choice than an LLM?

Links
AI in B2B Commerce, 17 November in Munich (in German): https://foobar.agency/de/events/ki-im-b2b-handel-muenchen-2026
Impact Check: https://foobar.agency/en/impact-check
Gartner on agent washing (June 2025): https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027
Gartner on agents built by vendor engineers (September 2026): https://www.gartner.com/en/newsroom/press-releases/2026-09-29-gartner-predicts-70-percent-of-enterprises-will-abandon-agentic-ai-built-by-vendor-forward-deployed-engineering-by-2028
Anthropic on token usage of multi-agent systems: https://www.anthropic.com/engineering/multi-agent-research-system

What Works Tomorrow. The podcast for decision-makers in retail and industry. By Matthias Dietrich and Philipp Krueger, Co-CEOs of foobar Agency.

#AI #AIAgents #AgenticAI #B2B #WhatWorksTomorrow