Agent washing: how to spot a real AI agent in 2026?

Third demo of the quarter for the finance and IT team at a Quebec distributor. The vendor presents "its AI agent". On screen, the same conversational assistant as last year, with a new name and the same limits.

The scene has repeated everywhere since 2025. Anything that touches an ERP or a financial process is now sold under the "agent" label. Sorting the real from the rebranded has become a buying skill in its own right.

“Over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls.” Source: Gartner, press release, June 25, 2025

What is agent washing, exactly?

Agent washing is the rebranding of an existing product, a virtual assistant, RPA or chatbot, as an "AI agent" without adding substantial autonomous capabilities. The term comes from Gartner, which documented it in June 2025 and estimates that only about 130 vendors, out of the thousands claiming the label, offer real agentic capabilities.

The nuance carries weight in an ERP. An assistant answers your questions. An agent detects a situation, decides within written rules, executes the action in the system, then escalates uncertain cases to a human.

Our review of AI agents in ERPs details what SAP, Microsoft and Oracle actually deliver in 2026. The contrast with some brochures is instructive.

Why will more than 40% of agentic AI projects be canceled?

Because costs climb, business value stays unclear and risk controls are missing, according to Gartner (2025). Agent washing makes all three worse. An organization that buys an "agent" and receives an assistant discovers the gap mid-deployment, exactly when workarounds cost the most.

MIT reaches a neighbouring conclusion by another route. According to The GenAI Divide study (NANDA initiative, 2025), 95% of generative AI pilot projects produced no measurable return. It is a preliminary study, and its methodology has been debated, but its finding lines up with what Gartner observes on agents.

Meanwhile, the pressure to invest stays high: 54% of CFOs make AI agent integration a priority, according to Deloitte (CFO Signals, fourth quarter of 2025). Buying under pressure, in a made-up market, leads straight to the canceled project. The causes are detailed in our analysis of why AI projects fail.

How do you tell a real AI agent from a rebranded product?

By verifying measurable capabilities rather than the product sheet: execution of a complete process, decisions bounded by written rules, learning from exceptions, an audit trail and a quantifiable result. A product that keeps these five promises holds up in a demonstration on your data, not only in a presentation.

A few benchmarks to situate the market in 2026:

• About 130 credible vendors out of thousands using the "agentic" label (Gartner, 2025)
• 19% of organizations invested significantly in agentic AI and 42% conservatively (Gartner poll, January 2025, 3,412 respondents)
• 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from less than 5% in 2025 (Gartner, August 2025)
• 95% of generative AI pilot projects produced no measurable return (MIT, The GenAI Divide, 2025)
• 19.2% of Canadian businesses were using AI in production in the second quarter of 2026 (Statistics Canada)

Methodological note: these figures come from public studies by Gartner, MIT and Statistics Canada. Scopes vary between studies, and none of these benchmarks describes your specific situation.

What questions should you ask a vendor before signing?

Five questions are enough, asked in this order, with proof required at every answer.

Show the complete process on our data, from the triggering event to the entry written in the system. A scripted demo on clean data proves nothing.

What does the agent do when it hits an exception? A real agent escalates the case with a clear reason. A rebranded product stops, or worse, improvises.

Who approves what, above which threshold, with what audit trail? The four controls to demand are detailed in our guide to AI agent governance in finance.

Which business indicator does the agent improve, measured against which baseline? Without a before number, no after number will mean anything.

And the last one: what remains of your product if the word "agent" is removed from the brochure?

If the product stays identical once the word "agent" is removed from the brochure, you have just met agent washing.

In the mandates PlanAxion carries out in accounts receivable, the most revealing test remains the benchmark: three months of your own payments, your remittance advices exactly as they arrive, and results measured at input and output. The cash application solutions PlanAxion implements go through that same evaluation format themselves.

Where should you start without getting caught?

With a high-friction process and a measurable indicator, never with the technology. Cash application, month-end close or dispute triage offer documented rules, high volumes and a result quantifiable in days and dollars. Exactly the terrain where a real agent proves itself.

That is the approach of our rapid AI solutions workshop: 4 weeks, 5 steps (prepare, identify, prioritize, validate data, decide and deliver), and every idea goes through the same filter of problem, value, data and effort. The investment varies with scope and is confirmed during a short exploratory call.

Will the sorting happen without you?

No. Gartner predicts over 40% of projects will be canceled, which means close to 60% will survive, and they will belong to the organizations that demanded proof before signing. The market is already settling. Buyers who test on their own data buy better, and often for less.

Frequently asked questions about agent washing

What is agent washing?

Agent washing is the practice of rebranding existing products, chatbots, RPA or virtual assistants, as "AI agents" without real autonomous capabilities. Gartner documented the phenomenon in 2025 and estimates that only about 130 vendors, out of thousands, offer true agentic capabilities.

What is the difference between an AI agent and a chatbot?

A chatbot answers questions and suggests actions. An AI agent pursues a goal: it detects a situation, decides within written rules, executes the action in your systems, then escalates uncertain cases to a human with a clear reason.

How do you test an AI agent before buying it?

Require a benchmark test on your own data: three months of real transactions, your documents exactly as they arrive, and results measured at input and output. Also ask for a demonstration of approval thresholds and the audit trail, two things a rebranded product can rarely show.

Why do agentic AI projects get canceled?

Gartner identifies three causes: escalating costs, unclear business value and inadequate risk controls. The firm predicts that over 40% of agentic AI projects will be canceled by the end of 2027. A bounded, measured and governed use case reduces each of these risks.