Agent washing (rebranding chatbots as AI agents)

The term was popularised by Anushree Verma, senior director analyst at Gartner, in a June 2025 press release. Verma cites a 40 percent cancellation rate for agentic AI projects by the end of 2027 as a consequence of two factors: agent washing on the supply side and FOMO on the demand side.

A test for agentic capabilities

A real AI agent has four technical traits that a chatbot lacks. First, it uses tools: it calls APIs, reads databases, writes to systems. Second, it plans across multiple steps: it breaks a goal into a sequence of steps, executes them, checks the results, and corrects. Third, it is observable: every decision is logged in a way that makes it possible to reconstruct why the agent did X. Fourth, it iterates based on feedback.

If a vendor cannot answer yes to three of the four, they are not selling an agent. They are selling a chatbot with better branding.

The financial scale

National CIO Review reports that global spending on GenAI consulting rose to 3.75 billion dollars in 2024, three times higher than the year before. A significant part of that figure goes to firms selling workflow automation with an added LLM layer as full agentic deployments. The client pays for a solution priced at a hundred thousand zloty and receives a solution worth thirty thousand.

Practical recommendation

Before signing a contract for agentic AI, it is worth requiring the vendor to run a live demo with three scenarios: a task that requires multi-step planning, a task that requires external tools, and a task with an error in the first step that the agent has to correct. A demo on cherry-picked examples from a slide deck is not enough.

We diagnose a company's readiness for agentic AI within the AI Readiness Audit, from five to eighteen thousand zloty, one to two weeks.