Agent Washing: How to Spot a Chatbot in a Trench Coat

Every category with real budget behind it eventually gets a wave of relabeling. Cloud-washing. Now agent-washing. If a term is worth paying more for, someone will apply it to something that doesn’t quite earn it, and the buyer is left doing the sorting.

Here’s the sorting.

What actually separates an agent from a chatbot

A chatbot, even a very good one, follows a decision tree. Given input A, it selects response B from a defined set of options. It can be dressed up with natural language and still be, structurally, a flowchart.

An agent reasons toward a goal across multiple steps, can choose from an open-ended set of actions rather than a fixed script, and can adapt its plan when the first attempt doesn’t work. That’s a meaningfully different piece of technology, and a meaningfully different price tag.

Five questions that expose the difference

Ask any vendor these, and the deflections are the answer:

1. “Walk me through what happens when the first approach fails.” A real agent adjusts its plan. A relabeled chatbot has a generic fallback message.
2. “What decision is it actually making, versus what response is it selecting from a list?” If every possible output was written by a human in advance, it’s a chatbot with a better vocabulary.
3. “Can it take an action outside the conversation, like updating a record or triggering a workflow?” Agents act. Chatbots talk.
4. “How does it handle a request nobody anticipated?” A scripted system breaks or deflects. An agent reasons, even imperfectly.
5. “What does it do when it’s uncertain?” This is the tell most vendors can’t answer well. A real agent has some mechanism for recognizing its own uncertainty. A rebadged chatbot just answers confidently either way.

Why this matters more than semantics

This isn’t a pedantic distinction for its own sake. Budget, governance, and risk exposure should look very different depending on the answer. A chatbot mistake is usually an annoying, contained failure. An agent mistake, at agent speed and agent scope, can be a much bigger one. Buying agent-level governance for a chatbot is wasted spend. Buying chatbot-level oversight for a real agent is a real exposure.

The buyer’s job

Not to distrust every vendor by default. To ask the five questions above before the contract is signed, and to size the governance investment to the honest answer, not the marketing deck.

Mar Dat helps enterprises evaluate what they’re actually buying before they buy it. Start with the free AI Readiness Assessment.