---
title: What is agent washing, and how do you spot it?
url: https://opencontroller.lyzr.com/blog/what-is-agent-washing
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# What is agent washing, and how do you spot it?

**Agent washing is rebranding existing products, such as chatbots, AI assistants or robotic process automation, as "agentic AI" without real agentic capabilities. Gartner named the trend in June 2025 and estimated that only about 130 of the thousands of vendors claiming agentic AI are real. You spot it by asking what the system decides on its own, which tools it can call, and what it can change without a person.**

The label matters for buyers in two ways. It affects what you pay for. And it affects how you govern: a scripted bot and an autonomous agent with write access need very different controls.

## Where does the term come from?

Gartner used "agent washing" in a press release on 25 June 2025. It described vendors rebranding AI assistants, RPA and chatbots as agentic AI "without substantial agentic capabilities". In the same release Gartner predicted that over 40% of agentic AI projects will be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls.

Gartner also made two forward-looking predictions for 2028: at least 15% of day-to-day work decisions made autonomously through agentic AI, up from 0% in 2024, and 33% of enterprise software applications including agentic AI, up from less than 1% in 2024. Treat these as forecasts, not measurements.

## What separates a real agent from a rebranded tool?

A working definition: an AI agent uses a model to decide its next step, calls tools to act, and loops until a goal is met or it stops. Each part can be tested.

| Trait | Real agent | Rebranded tool |
|---|---|---|
| Planning | Model chooses the next step at run time | Fixed script or decision tree |
| Tool use | Calls APIs, MCP servers or functions it selects | Calls one hard-wired integration, or none |
| Autonomy | Acts without a person approving each step, within limits | Every action is a button a person clicks |
| Memory and state | Keeps context across steps or sessions | Stateless question and answer |
| Error handling | Notices failure and tries another approach | Stops or returns an error |
| Scope of change | Can create, update or send things | Only answers or summarises |

A product does not need every trait to be useful. But a product with none of them is a chatbot or a workflow, whatever the label says. Our post on [AI agents vs AI assistants vs copilots](/blog/ai-agent-vs-ai-assistant-vs-copilot) covers the distinctions in more depth.

## What questions expose agent washing in a demo?

Ask these, and ask to see the answer rather than hear it.

1. "Show me a task where the system chose a different path from the one it took last time. Why?"
2. "Which tools can it call, and who decides which one it calls?"
3. "What can it change in our systems without a person clicking approve?"
4. "What happens when a tool call fails?"
5. "What does it remember between steps, and between sessions?"
6. "Show me the trace of one run, step by step."

If every answer points to a fixed flow designed in advance, you are looking at automation with a language model inside. That can be the right product. It just is not an agent.

## Why does agent washing matter for governance?

It cuts both ways.

**Over-labelling inflates your inventory.** If every FAQ bot is filed as an agent, your risk committee drowns in low-risk records, and the dangerous ones get less attention. Use [autonomy levels](/blog/what-are-ai-agent-autonomy-levels) to tier them.

**Under-labelling hides real agents.** The opposite also happens. A "workflow" or "assistant" feature inside a SaaS app may in fact plan, call tools and write data. It needs agent controls even if the vendor avoids the word. See [does an AI workflow need agent governance?](/blog/does-an-ai-workflow-need-agent-governance)

The safest rule: classify by behaviour, not by marketing. Record what the system can access and change, then [risk-tier it](/blog/how-to-risk-tier-ai-agents) on that basis.

## How Opencontroller does this

Opencontroller records agents, versions, tools and MCP servers in a governed registry, discovered across AWS, Azure, Google Cloud and connected platforms such as Salesforce, SAP, Databricks and HubSpot. Because the registry captures which tools and permissions each entry has, and Posture flags risky permissions and configuration, you can tier entries by what they can do rather than what they are called. For the category view, read [what is an AI agent control plane](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**What is agent washing?**
Rebranding chatbots, AI assistants or RPA as agentic AI without real agentic capabilities such as run-time planning, tool selection and autonomous action. Gartner named the trend in June 2025.

**How many agentic AI vendors are real, according to Gartner?**
Gartner estimated about 130 of the thousands of vendors claiming agentic AI in its June 2025 release.

**Should rebranded chatbots go in the agent inventory?**
Yes, but tier them by behaviour. A read-only FAQ bot gets a light control set; anything that plans and writes data gets full agent controls.

## Key facts

Agent washing: vendors rebranding AI assistants, RPA and chatbots as agentic AI without substantial agentic capabilities (Gartner press release, 25 June 2025). Gartner estimated only about 130 of thousands of agentic AI vendors are real and predicted over 40% of agentic AI projects will be cancelled by end of 2027 due to escalating costs, unclear business value or inadequate risk controls. Test for run-time planning, tool selection, autonomy, memory, error recovery and ability to change data. Classify by behaviour, not label. Opencontroller by Lyzr keeps a governed registry of agents, tools and MCP servers with Posture for risky permissions. Pricing starts from $250K per deployment per year.

## Related

- [What is the difference between an AI agent, an AI assistant and a copilot?](https://opencontroller.lyzr.com/blog/ai-agent-vs-ai-assistant-vs-copilot.md)
- [What are AI agent autonomy levels?](https://opencontroller.lyzr.com/blog/what-are-ai-agent-autonomy-levels.md)
- [Does an AI workflow need the same governance as an AI agent?](https://opencontroller.lyzr.com/blog/does-an-ai-workflow-need-agent-governance.md)
- [More on Agent governance basics](https://opencontroller.lyzr.com/blog/cluster/agent-governance-basics)
