Agent washing is the practice of marketing an existing chatbot, assistant, or rule-based automation as an autonomous AI agent without the underlying capability to back it up. Gartner coined the term in June 2025 and estimated that of the thousands of vendors claiming agentic AI, only about 130 were genuinely building it.

That figure is worth treating as directional rather than exact. Gartner has not published a public methodology for how it counted the denominator. But the direction of the finding, that the category is dominated by relabeled products rather than new capability, is echoed across the industry and matches what buyers report when they actually test a product against its claims.

What agent washing looks like in practice

Agent washing is not always dishonest. There is no certification body for what counts as an AI agent, no agreed threshold, no standards test. In that vacuum, most vendors set the bar where their product already stands.

The pattern shows up in a few consistent ways:

• A chatbot with a new system prompt and a rebranded landing page

• Robotic process automation, following fixed rules, marketed as autonomous decision-making

• A product that can call one API in a demo, presented as full workflow execution

• Marketing copy that uses "agent," "autonomous," and "AI-powered" interchangeably, without defining any of them

None of these are agents by the functional definition covered in AI Agent vs. Chatbot: What's the Difference: a system that perceives context, plans a sequence of steps, uses tools to act, and carries memory across that sequence without a new prompt at every turn.

Why the label runs ahead of the capability

Vendor incentives explain most of the gap. "Agentic" commands a pricing premium and a faster sales cycle right now, and the term is loose enough that almost any product can be repositioned to fit it without changing a line of code.

The cost of that gap is not abstract. Gartner projects that more than 40% of agentic AI projects will be canceled by the end of 2027, attributing the failures to escalating costs, unclear business value, and inadequate risk controls, not to weak underlying models. Separately, MIT's Project NANDA found that 95% of enterprise generative AI pilots show no measurable financial return, with only a small minority of pilots reaching production and delivering real value.

Both findings point the same direction: the failure usually happens after the sale, in the gap between what was demoed and what the product can actually sustain in a live environment. That gap is exactly what separates a vendor's technology from the work of an implementation partner, and it is a large part of why agentic AI projects fail even when the underlying model is sound.

A buyer's checklist for spotting agent washing

Run any vendor claiming "AI agent" through these five questions before a contract, not after.

1. What specific action does it take without a human doing the last step? Ask for a concrete example, not a description. A real agent finishes a task; a washed product hands the final step back to a person.
   
   

2. What happens when a step fails? Autonomy includes handling errors and adjusting course. If the only answer is "a person gets notified," the system is not planning or adjusting, it is alerting.
   
   

3. How many systems can it actually call today, not on the roadmap? Tool use is a defining trait of an agent. A demo built around one hardcoded integration is not evidence of general tool use.
   
   

4. Does it carry context across steps, or does every step start fresh? Memory across a multi-step process is what separates an agent from a chain of disconnected chatbot calls.
   
   

5. Can the vendor name the team that will deploy this inside your environment, separate from the sales team? A real deployment requires implementation work most vendors cannot do themselves at scale, which is where a verified implementation partner becomes part of the evaluation, not an afterthought.

A vendor that answers all five with specifics is worth taking seriously. A vendor that answers with adjectives, "seamless," "intelligent," "next-generation," instead of mechanics is showing the tell.

FAQ

What is agent washing?
Agent washing is the practice of marketing a chatbot, assistant, or rule-based automation as an autonomous AI agent without the planning, tool use, and multi-step autonomy that define one. Gartner named the practice in 2025.

How many AI agent vendors are actually real?
Gartner estimated in June 2025 that of the thousands of vendors claiming agentic AI, only about 130 were genuinely building agentic capability. Gartner has not disclosed a full methodology for that count, so the figure should be read as directional.

How do I know if a vendor is agent washing their product?
Ask for a specific example of the product completing a task end to end without a human doing the final step, how it handles a failed step, and how many live integrations it actually uses today. Vague answers built around adjectives rather than mechanics are the clearest signal.

Why does agent washing matter if the product still works reasonably well?
A relabeled chatbot might still be useful, but buying it expecting autonomous, multi-step execution sets a project up to fail on scope it was never built to handle. Gartner attributes more than 40% of agentic AI project cancellations by 2027 to exactly this kind of mismatch between claim and capability.

Bonobee is the discovery, matching, and verification infrastructure for the agentic AI ecosystem. Verified profiles document what a vendor or implementation partner has actually deployed, not what they claim in a demo. Search Bonobee's directory, or read more on why the agentic AI ecosystem needs a verification layer.