What is agent washing?
What is the difference between a chatbot and an agent?
A chatbot answers when asked. A script, such as robotic process automation (RPA), repeats fixed steps.
An agent, in the sense vendors imply, plans steps toward a goal and acts without being told each one. That is the claim. Agent washing is when the claim is made and the product is a chatbot or a script.
Nothing is wrong with a good chatbot or a good script. The problem is paying for autonomy you did not get, or trusting autonomy that is not there.
How do you spot it?
Ask what the product does on its own, step by step. Then ask where a person approves.
If the answer is vague, treat the label as marketing. If the answer is a list of fixed steps, you are looking at automation. Both can be useful. Neither is an agent in the sense of the pitch.
Check also what the tool can reach. A connector through the Model Context Protocol gives access, which is not the same as acting on its own. And any tool that can act on your accounts needs a human in the loop.
For ZFD: Our view is that "agentic" is a description to test, not a feature to buy. For what AI does well and badly with marketing data, read what an AI analyst gets right and wrong.
Easily confused with
Agent washing vs Greenwashing
The same move in another field: a claim that outruns the product.
Agent washing vs A real agent with limits
A tool can be genuinely agentic and still need approvals. Limits are not proof of washing.
Buzzword alert
"Our autonomous AI agent runs your marketing"
Usually means: a chatbot or a set of scheduled automations with a new name.
Ask: what did this product do before it was called an agent?
"Agentic workflows built in"
Usually means: a fixed sequence of steps with a language model in one of them.
Ask: what happens when a step fails?
"Fully agentic reporting"
Usually means: a summary generated from a dashboard.
Ask: can it explain how each number was calculated?
FAQ
Is agent washing a real term or a joke?
It is used seriously. The press report we cite attributes it to Gartner1.
What did Gartner say about agentic AI projects?
Trade press coverage of a Gartner forecast from June 2025 says a large share of agentic AI projects could be cancelled by the end of 2027, and names agent washing as part of the hype1. It is a forecast, not a measurement.
Are all AI agents fake?
No. The point is to test the claim. A real agent can describe its steps and where a person approves.
Sources
Let’s find what’s actually working.
Next step. Been pitched an AI tool and not sure what it does? Talk it through with us.