What is agent washing?
Rebranding chatbots and RPA as AI agents without real autonomy: Gartner estimates only ~130 truly agentic vendors out of the thousands claiming it.
Agent washing is the practice of reselling products that are not agents as "AI agents": chatbots, assistants and RPA with a rebranding, and no real autonomy in deciding and acting. The term, modeled on greenwashing, received Gartner's stamp: in a June 2025 press release it estimated that, out of the thousands of vendors claiming to be "agentic", only about 130 offer real agentic capabilities. The phenomenon comes down to a simple incentive: the label "agent" opens budgets today and justifies prices that "chatbot" no longer does, the same dynamic already seen with cloud washing and AI washing more broadly. The buyer ends up footing the bill: a project built on a rebranded chatbot sold as an agent almost always fails, because it is asked to do something it simply cannot, and the gap only becomes visible after the contract is signed.
Why it happens (and why it concerns you)
The incentive is trivial: "agent" opens budgets today and justifies prices that "chatbot" no longer does. It is the same dynamic already seen with cloud washing and AI washing: when a label becomes valuable, it ends up on everything. The cost, though, lands on the buyer: in the same release Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, and part of the problem is precisely starting from expectations built on products sold as something they are not. An "agentic" project built on a rebranded chatbot fails by definition: you asked it for something it cannot do.
The buyer-side checklist
Five questions dismantle most of the washing. One: does the system decide on its own the steps to reach a goal, or does it execute a flow predefined by a human? Two: does it act on real systems (writing data, calling APIs, performing operations) or does it only answer questions? Three: does it complete a multi-step task without you guiding it step by step? Four: what happens when it gets things wrong, who supervises it, what logs does it leave? Five: does the vendor accept a demo on your data and your processes, rather than on its prepared examples? And an honest note to close: if the true answer is "it is a chatbot with a new name", that is not necessarily a tragedy. Chatbots do plenty of jobs well. Just buy them, and pay for them, as chatbots.
Related terms
- AI Agents (Agentic AI) · AI systems that go beyond answering: they plan, use tools and take actions autonomously inside your processes.
- AI Employees / Digital Workers · A marketing label for an AI agent presented as a digital employee with a name and role: almost always the same agentic AI technology, not a different leap.
- Guardian agents · AI agents that supervise other agents: they monitor, control and, when needed, contain. Gartner sees them at 10-15% of the agentic AI market by 2030.
- Service-as-Software (SaS) · SaaS inverted: you no longer buy a tool to do the work, you buy the work done. AI sells outcomes, not licenses.
- AI project failure causes · The recurring causes behind AI project failure: no owner, no agreed metric, unready data, no process redesign.
A term that hits close to home? Let's talk.
CONTACT ME