Agent washing
Agent washing is the practice of presenting a tool as an AI agent when it's really a chatbot, a simple automation, or a flow with little real autonomy. The term warns against the inflated marketing use of "agent" without enough evidence of planning, tool use, execution, adaptation, and control.
Agent washing appears when the market adopts a buzzword and many products start using it even though their actual capabilities haven't changed substantially.
Not every AI system is an agent. And not every agent delivers useful autonomy.
What characterizes a real agent
An AI agent usually combines several elements:
- a goal or task;
- instructions;
- context;
- the ability to execute steps;
- tool use;
- memory or state;
- verification sensors;
- limits and permissions;
- human supervision.
A tool that only answers a question or runs a fixed flow shouldn't be sold as an autonomous agent.
Signs of agent washing
- The product used to be called an assistant and is now called an agent, with no real changes.
- It's unclear which decisions it makes autonomously.
- It doesn't use external tools or maintain state.
- It requires constant human intervention but is marketed as autonomous.
- It offers no traceability of actions.
- It doesn't allow limits or permissions to be configured.
- It shows no effectiveness metrics.
- It confuses automation with agency.
Why it matters
Agent washing creates false expectations. It can lead to misdirected investment, operational risks, and team frustration.
In agile teams, the problem isn't just buying a bad tool. The problem is designing processes around an autonomy that doesn't exist.
How to evaluate it
Before adopting an "agentic" tool, it's worth asking:
- Which tasks can it complete end to end?
- Which tools can it use?
- Which decisions does it make, and which does it not?
- How are its outputs reviewed?
- What permissions does it need?
- What happens if it gets something wrong?
- How are its actions audited?
- What evidence is there of improvement over an ordinary assistant?
Common mistake
Buying the label instead of assessing the capability. A tool presenting itself as an agent doesn't mean it can act with useful autonomy. The team should review real features, limits, traceability, controls, and evidence before integrating it into their working system.
Resources
🏦 Scrum in teams with AISkill Arena · Scrum Manager
References
- Reuters. (2025). "Over 40% of agentic AI projects will be scrapped by 2027, Gartner says." Reuters.
- Writer. (2025). "Enterprise guide to agent washing." Writer.
See also
Want to take your agility further? You can look for upcoming course and exam dates, or go at your own pace by becoming an Agile Club member. Membership includes exclusive resources, e-learning classrooms, and access to Skill Arena: a space to practice and measure your agile skills whenever it suits you.