What is Agent Washing?

Q: "How can I tell if a product is agent-washed?"

"Ask the vendor what happens when the system encounters something unexpected mid-task. If it cannot change its plan based on intermediate results, if it follows a fixed sequence regardless of what it finds, it is automation with a language model interface, not an agent."

Q: "Is agent washing the same as AI washing?"

"Agent washing is a specific strain of AI washing. AI washing broadly covers overstating AI capabilities in any product. Agent washing specifically targets the agentic AI category, where vendors relabel existing automation as autonomous agents."

The practice of marketing chatbots, workflow automation, or RPA tools as AI agents to capitalize on hype, even when the product lacks genuine autonomous reasoning or adaptive behavior.

When every vendor in the market starts calling their product an “AI agent,” the term stops meaning anything. That is where we are.

Agent washing is the AI-era sequel to AI washing. Where AI washing meant slapping “powered by AI” on a rules-based system, agent washing means calling a chatbot or workflow automation tool an “agent” because the word commands premium pricing and executive attention. The underlying product has not changed. The positioning has.

How it works

A vendor takes an existing product, a chatbot that answers questions from a knowledge base, a workflow that runs a fixed sequence of steps, or an RPA bot that clicks through screens, wraps it in language model integration, and markets it as an AI agent. The product may use a language model for natural language understanding, but it cannot reason about its next step, adjust its plan based on intermediate results, or select tools dynamically. It follows a predetermined path with a conversational interface on top.

Why it matters

Agent washing inflates expectations and distorts purchasing decisions. Organizations invest in “agentic” products expecting autonomous task completion and get a chatbot that requires the same manual orchestration as before. The resulting disappointment erodes trust in AI investments broadly, not only in the vendor that over-promised.

The filter

Ask one question: what does the system do when something unexpected happens during execution? If it stops, errors out, or continues on its fixed path regardless, it is not an agent. Genuine agents evaluate the unexpected result and adjust their approach. That adaptive reasoning at runtime is the line between agent and automation.

Frequently Asked Questions

How can I tell if a product is agent-washed?

Ask the vendor what happens when the system encounters something unexpected mid-task. If it cannot change its plan based on intermediate results, if it follows a fixed sequence regardless of what it finds, it is automation with a language model interface, not an agent.

Is agent washing the same as AI washing?

Agent washing is a specific strain of AI washing. AI washing broadly covers overstating AI capabilities in any product. Agent washing specifically targets the agentic AI category, where vendors relabel existing automation as autonomous agents.