There is no denying we live in an agentic era. But first, what is Agentic? Merriam-Webster defines Agentic as having the capacity to act independently, make decisions, and take initiative to reach a specific goal. Agentic, which is now commonly used to imply automation, is not the same as classic automation. Now, let’s look at the history of automation.
Little Bit of History
Humans have always aspired to automate mundane tasks. Again, according to Merriam-Webster, Automation is defined as: the operation of a device, process, or system by mechanical or electronic devices that replace a human operator. Little history: the first use of the word Automation is generally attributed to Ford Motors executive Delmar S. Harder, who was the president of the manufacturing division in the post-WWII era (1947-48). According to the third volume of a history of Ford by Allan Nevins and Frank Ernest Hill (Ford: Decline and Rebirth, 1933-1962 [New York, 1963], p. 354), during a staff meeting focused on the study of efficient handling of materials, Harder coined up a new departmental name called ‘Automation Department’.
Historically, automation aims to speed up work, reduce mistakes, and handle repetitive tasks so humans can focus on tasks that cannot be automated. One implied outcome of automation is improved safety, especially for tasks that are dangerous for humans. Thus, automation leads to idempotency. An idempotent operation is a process that can be applied multiple times and get the exact same system state as when applied just once. Mathematically - (f(f(x)) = f(x)). This is the essence of deterministic outcomes – to enhance predictability. This is where Agentic starts to deviate from the implicit outcome of automation.
Automation to Agentic
We are firmly in the Agentic era, and we have gradually moved away from simply automating mundane tasks toward developing agents that can reason and operate independently like humans. This is the trade-off of Agentic automation, which brings both opportunities and challenges. How can we ascertain predictable outcomes from non-deterministic processing? This is why Agentic Governance has become a hot topic for agentic processing.
Consider the following guidelines for your automation strategy:
-
Define the problem and outcome criteria - Many projects tend to skip this step by superficially thinking of the outcome and making a predetermination that Agentic is the way to go. Yet most problems can be solved with classic automation based on simple decision trees that can be codified, requiring no reasoning. This can save token burn.
-
Do the outcomes require idempotency, meaning deterministic outcomes? From #1, if the problem and outcome evaluation needs predictability, then the processing may require idempotency.
-
Ensure the starting point, or bias, is simple automation, then agentic. If the outcome requires idempotency, classify the automation as simple. If idempotency is not a requirement, classify as Agentic automation.
-
Classify the project by creating a grid showing which problems fall into simple automation versus agentic automation and set to inform your broader automation strategy.
These guidelines will help you understand the cost implications, acceptable criteria for idempotency, and the sustainability of the workflows.
In closing, before you embark on an Agentic/Automation strategy, follow strict guidelines to know where simple automation ends and agentic work begins. Consider the outcomes in alignment with idempotency, cost implications, and the sustainability/supportability of the automation and agentic infrastructure.