Over the past year, the question asked in automation meetings has changed. It used to be: "Can we hand this process to a robot?" Now it is: "Has RPA become obsolete, or should we move straight to an AI agent?" With "agent" now added to the name of nearly every product in vendor presentations, the decision is getting harder, not easier.
This article treats the two technologies not as rivals but as tools for different jobs. The aim is to choose the right tool process by process, and to make a habit of asking what lies behind the "agent" label.
"Agent washing": the label changed, the product didn't
In an assessment published in June 2025, Gartner issued a warning. Many vendors are rebranding their existing products without any significant agent capability: AI assistants, robotic process automation (RPA) and chatbots. Gartner calls this "agent washing". It estimates that only about 130 of the thousands of "agent" vendors are real.
The same assessment contains two more forecasts. More than 40% of agent projects will be canceled by the end of 2027. The reasons are rising costs, unclear business value and inadequate risk controls. In a Gartner poll of 3,412 webinar attendees conducted in January 2025, 19% of organizations say they have made significant investments in agents. Another 42% have invested conservatively, 8% have not invested at all, and 31% are waiting or undecided.
The second assessment, from August 2025, shows another side of the conceptual confusion. According to Gartner, the most common misconception is labeling AI assistants that depend on user input and do not work independently as "agents". An assistant makes the user's work easier. An agent, by contrast, can carry out a task end to end, assess the situation and take action.
Three tools, three different jobs
Gartner's recommendation is really a simple division of labor: use agents when decisions are required, automation for routine workflows, and assistants for simple information retrieval. This split can be turned into a process-by-process choice:
| Nature of the work | Suitable tool | Typical example |
|---|---|---|
| Clear rules, fixed steps, structured input | RPA / classic workflow automation | Data transfer between systems, standard report generation, document archiving |
| Finding information quickly or producing a text draft | AI assistant | Searching a policy document, drafting an email, summarizing a meeting |
| Assessing variable input and carrying out multiple steps | AI agent | Classifying an incoming request and starting a transaction in the relevant system, detecting an exception and routing it |
The critical question here is this: is there really a decision point in the process? In Gartner's words, many use cases positioned as "agents" today do not actually require an agent. Building an agent for a process whose rules can be written down up front means choosing a path that is more expensive and harder to audit.
Five questions to ask before choosing
1. Can the process be described with rules? If you can lay out the steps in a flowchart completely, RPA or workflow automation is often enough. If the chart says "it depends" at some step, that step is a decision point and can be an agent candidate.
2. How variable is the input? Classic automation gives reliable results on data that arrives in the same format. If there is free text, documents in different formats or unpredictable requests, the capacity to interpret is required.
3. What does an error cost? The more an agent decides on its own, the more important risk control becomes. Inadequate risk controls are also among Gartner's reasons for cancellation. In processes where an error is costly, a human approval step should be part of the design from the start.
4. How will value be measured? Gartner recommends using agents only where they deliver clear value or return on investment. Before the project starts, a metric such as time, error rate or transaction volume should be chosen, along with its starting value.
5. How will it connect to existing systems? According to Gartner, integrating agents into legacy systems can be technically complex. It can disrupt workflows and require costly changes. In most cases, Gartner sees redesigning the workflow from scratch around the agent as the ideal path.
Questions to ask the vendor
When evaluating a product that carries the "agent" label, these questions help you see what lies behind the label:
- Can the product carry out a task end to end without the user giving a command at every step, or does it wait for human input at each step?
- At the points where it makes a decision, what information does it rely on, and can that decision be logged and reviewed afterward?
- How are the systems the agent can act in, and the permissions it acts with, limited?
- Why can't the same job be done with an existing RPA or workflow tool?
- By what criterion will success be judged at the end of the pilot?
The last two questions are especially important. If the answer is "the same job can also be done with rule-based automation", the cost and complexity of an agent may not be justified. If no clear answer emerges, it is also uncertain whether the product is a real agent.
RPA isn't dead; its role is becoming clearer
According to Gartner's August 2025 forecast, 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026. That share was below 5% in 2025. In other words, agents are quickly moving into the software that organizations use. This does not mean that rule-based automation is disappearing. Gartner positions the value of agents as "beyond the capabilities of coded automation bots and virtual assistants". In other words, the agent is establishing itself not in place of automation but in the decision-requiring work that automation cannot reach.
The practical takeaway is this: build the automation portfolio by the nature of the processes, not by technology. Processes with clear rules stay with rule-based automation. Processes that contain a decision point, have measurable value and have risk controls designed for them become agent candidates. An "agent" decision made without this distinction risks ending up among the projects Gartner predicts will be canceled.
Why logistics experience matters here. Logistics operations contain both kinds of work this article separates. Moving order data between systems, producing standard stock reports and archiving delivery documents are rule-based tasks with structured input, the first row of the table. A delayed or damaged shipment reported in free text, which has to be read, matched to an order and routed, is a genuine decision point and an agent candidate. Because a wrong decision in a warehouse or on the road sends real goods to the wrong place, the cost of an error (question 3) is typically managed there through approval steps and written exception rules, and because logistics teams usually track cycle time, error rate and volume, the baseline that question 4 asks for often already exists. One of our consultants brings more than 30 years in operations, logistics and supply chain: at a technology distribution company of roughly one billion dollars in scale, he ran logistics, warehousing, technical service and administrative operations end to end, building and leading teams of 100+ people and delivering lasting gains in operational performance through KPI-driven programs. He has also managed nationwide warehouse and transport operations at leading retail, logistics and distribution organizations. An industrial engineer, he now brings that supply-chain discipline to SMEs through AI and automation. Meet our consultants.
If you would like to read further on this topic, you may also look at our articles on the cancellation risk of AI agents, how widespread AI-powered RPA is in Europe and the access permissions granted to agents.
#interimmanagement #forwarddeployedengineering