Agent Washing

By analogy with greenwashing, agent washing describes a gap between promise and substance. A genuine agent accepts a goal, forms a plan, selects tools, evaluates intermediate results and adapts the plan. Many products sold as agents fulfil only the prompting part. For buyers this creates a concrete risk: budgets are released against autonomy promises that turn into manual work in operations. The difference becomes testable through a few questions — who defines the steps? Can the system pick tools not in the script? How does it detect its own failure? And which metrics evidence tasks completed without intervention?

For marketing decision-makers agent washing cuts both ways: in purchasing it risks misinvestment, in your own communication it risks challengeable claims. Precise, evidenced wording has become a competitive advantage rather than a brake.

A vendor advertises an "autonomous campaign agent". The demo reveals a fixed sequence of three prompts, no tool selection and no failure reporting. It is a workflow with LLM steps, not an agent.

The term was coined in 2025 by analyst firms estimating that only a small fraction of vendors marketed as agentic actually meet the criteria. In 2026 agent washing entered procurement guidance and the debate about advertising claims for AI features. In parallel, consumer and competition authorities began scrutinising exaggerated AI claims more closely, carrying the term from jargon into compliance practice.