A term coined by Gartner to describe products sold as "AI agents" that are in practice just conditional logic (if/else) with a superficial LLM layer.
Detailed explanation
Agent washing is the AI equivalent of greenwashing: products that sell themselves as autonomous AI agents but are in practice traditional deterministic systems with an LLM interface on top. Gartner coined the term to alert corporate buyers about the risk of investing in "agents" that lack real autonomous decision-making, goal iteration, or multi-tool usage capabilities. For product managers, agent washing is a red flag in vendor due diligence and in their own product communication: selling something as an "agent" when it is traditional automation can create unmet expectations and erode trust.
How to use it in product decisions
In product decisions, use agent washing as an honesty test for the proposition. Ask for evidence of autonomy, memory, tool use, and recovery after failure. Compare the claimed workflow with real logs and identify where deterministic rules do the work. The metric is not how often the word agent appears, but how many steps the system completes with quality, clear limits, and known human intervention. If the product cannot support that evidence, describe it as assisted automation.