Agent Washing
In 2026, everything is an AI agent.
Your email app has an agent. Your CRM has an agent. Your note-taking app, your customer support tool, your calendar. Some of these genuinely are agents: software that takes actions on its own to get a task done. A lot of them are the same chatbot they were last year, with a new label.
There's a name for that: agent washing.
What agent washing means
Agent washing is when a company markets a regular chatbot, a simple automation, or a rule-based script as an "AI agent" because "agent" is the hot word of the year.
It's borrowed from greenwashing, when companies make products sound more eco-friendly than they are. Same idea: the label promises more than the product delivers.
The term has been circulating in analyst and industry circles as agent products multiplied this year. Ironically, most of what you'll find if you search for it right now comes from companies selling agents, each explaining that other companies are agent washing.
So here's a neutral way to check for yourself.
What a real agent does
A real AI agent, at minimum:
- Takes actions, not just answers. It clicks, submits, sends, books, updates.
- Works through multiple steps on its own toward a goal.
- Adjusts when something goes wrong, instead of stopping at the first error.
- Uses tools or other apps to get the job done.
A chatbot answers. An automation follows a fixed script. An agent pursues a goal.
The 6-question agent washing test
Before you pay for anything sold as an "agent," ask these.
1. What does it actually do?
Ask for a concrete example: "Show me a task it completes from start to finish without me." If every example ends with it suggesting something for you to do, it's an assistant, not an agent.
2. What happens when a step fails?
A website is down. A form field changed. An email bounced. Does it try another route, or does it stop and throw an error? Rigid scripts break. Agents adapt, at least somewhat.
3. Can it handle a task it wasn't specifically built for?
Automations do exactly what they were configured to do. Agents can take a new instruction in plain language and figure out the steps. Try asking for something slightly unusual.
4. Which apps can it act in?
An agent that can only work inside its own product is limited. Real agents typically connect to other tools: email, calendars, browsers, databases.
5. Do you have to click "go" at every step?
Asking for approval before important actions is good design, and you want that for payments or sending emails. But if you're approving every single tiny step, you're doing the work and it's just typing for you.
6. What did they call it last year?
Look it up. If last year's version was the "AI assistant" and nothing meaningful changed except the name, that's your answer.
Why it matters
You might overpay. "Agent" is being used to justify higher prices. If the product is really a chatbot, compare it to chatbot prices.
You might get false confidence. If you believe a tool will handle a process on its own, you might stop checking it. If it's actually a brittle script, things break silently.
It muddies the real progress. Genuine agents are one of the biggest shifts in AI this year. Agent washing makes it harder to tell the real ones apart.
To be fair to companies
The line between "assistant" and "agent" genuinely is blurry, and there's no official definition. A tool can be partly agentic: good at one multi-step task, useless at others. That's fine, as long as the marketing is honest about it.
The problem isn't small agents. It's chatbots in agent costumes.
FAQ
What is agent washing?
Marketing a chatbot, simple automation or rule-based tool as an "AI agent" when it doesn't independently take actions toward a goal.
What's the difference between an AI agent and a chatbot?
A chatbot responds to your messages. An agent takes actions across multiple steps, often using other apps, to complete a task.
Where does the term agent washing come from?
It's modeled on "greenwashing" and became common in the tech industry in 2025-2026 as many products rebranded as agents.
How can I tell if a product is a real AI agent?
Ask it to complete a real multi-step task on its own, see how it handles failures, and check whether it can act in other apps.