When a vendor markets a chatbot, script, or rules-based automation tool as an “AI agent” despite it having no real autonomy, judgment, or memory.
Agent washing is the practice of marketing a product as an “AI agent” when it is really a chatbot, a script, or a robotic process automation tool with a new label. Gartner called it out by name in 2026, warning buyers across categories including supply chain and marketing software that many “agent” claims do not hold up under scrutiny.
A genuine AI agent sets its own sub-goals, chooses which tools to call, and adapts when a step fails. A washed version follows a fixed decision tree: if X happens, do Y. It might use a language model to write the response, but the model is not actually deciding what happens next. The tell usually shows up in the failure mode: ask it to handle something outside its script and a real agent improvises, while a washed one breaks or falls back to a canned reply.
Agencies sit on both sides of this problem. They buy AI tools to run campaigns, and increasingly they sell AI-powered services to clients, so agent washing cuts both ways.
Vendor evaluation gets harder. When every media buying platform, creative tool, and reporting dashboard claims agentic capability, procurement needs a way to separate the ones that actually plan and act from the ones that just added a chat interface.
Client trust is on the line. An agency that oversells its own AI capabilities, calling a scripted workflow an “agent,” risks the same credibility hit it would take for overselling media results. Gartner’s prediction that over 40 percent of agentic AI projects get canceled by 2027 traces partly back to this gap between claim and capability.
Pricing depends on getting this right. Agencies charging a premium for “agentic” services need to explain, concretely, what the system decides on its own versus what still runs on fixed rules, especially once a client’s procurement team starts asking.
A mid-size agency is evaluating three AI agent platforms for automated paid social optimization before renewing a media tooling contract. Two of the three turn out to be rules engines: if CPA rises above a threshold, pause the ad set, a workflow the agency’s own analysts already built in a spreadsheet years ago. The third actually reallocates budget across audiences it was not explicitly told to test, and explains its reasoning in a log the account team can review. The agency builds a short scorecard, does it choose its own next action, does it use tools beyond what it was told to use, does it recover from an unexpected result, and runs it against every vendor pitch from then on, catching two more washed tools the next quarter.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.