An old Chinese proverb says that every journey of a thousand miles starts with a single step.
When starting out, adopting agentic AI can seem like a journey of a thousand miles, at least as you try to keep up with the latest and greatest industry innovations.
Some agentic AI promoters go to the extreme, asking you to believe that AI agents will replace human employees and act without supervision to carry out any task you can think of.
Others in the industry are more pragmatic, drawing a clear distinction between traditional deterministic and modern probabilistic agentic outcomes, focusing on what each approach is good for.
Traditional IT systems – that is, what everyone used before the advent of generative AI – are deterministic. They produce precise and repeatable results because they use binary computer code with a single mathematical interpretation.
Generative AI systems, on the other hand – including agentic AI – are probabilistic because they use statistical matches (instead of exact matches) to generate responses to human language prompts.
It therefore makes sense to take the first step in working with agentic AI by picking a use case that clearly fits in the “non deterministic” or “probabilistic” category, such as a research project, analyzing and summarizing data, or requires context-awareness, rather than trying to replace a deterministic process.
Picking a Good Use Case
Instead of looking for a traditional deterministic use case such as using an IF statement to send a message (i.e. if a specific database error occurs, then send a specific message to the DBA), think about it in more human terms (i.e. let the DBA know when any error occurs).
In other words, instead of thinking about an agentic computing problem in terms of a mathematical comparison, think about an agentic computing problem in terms of a natural language prompt.
Instead of telling the agent how to figure out how to do it, let the agent figure it out, decide which data sources to use, which MCP servers or APIs to call, how to format the messages, and so on.
A lot of agentic AI behavior is determined by the data and tools they have access to. Leverage the flexibility to see how work can be automated in new ways.
Consider an agent that compares insurance policies to determine which is the best fit for a customer – something that is important but time-consuming to do.
Or an agent that recommends new products based on past purchases. Or one that finds similar items when an item the customer picks is out of stock.
The value of a non-deterministic AI agent derives from their ability to deal with ambiguity, interpret a set of instructions, and generate new artifacts to complete a complex task.
Take the Next Step
Once you have taken the first step and identified a good use case for an AI agent, the next step is to find a new approach to an old problem – using the power of generative AI to automate something you couldn’t do before. This could involve multiple agents, or an agent deployed in a multi-step workflow.
A number of use cases that could not be previously solved economically are now possible due to new applications for AI agents. Let’s see if we can find some.
For example, you could assign a task to an AI agent that includes having the agent figure out how to complete the task on its own.
In other words, the instructions could include how to decide something, such as determining whether to flag a security incident as significant or not, and fire off an alert to the security analyst with the right skills to respond.
In making decisions, AI agents take into account data from various sources and navigate a workflow, whose execution is improved by allowing an agent to make dynamic and potentially autonomous choices and actions.
At this level we’re talking about potentially assigning agents tasks we’d normally only assign to people, and training them on business rules and contexts that help people become more productive.
So also think about placing an agent within a larger context, and give it access to a Model Context Protocol (MCP) server to access the data and integration tools the agent needs to discover how to dynamically execute a task.
How Boomi Helps
The Boomi Enterprise Platform combines non-deterministic AI agents within deterministic workflows so you can start with a deterministic task and add however much non-deterministic work you want to include within your enterprise applications.
The platform features Boomi Agentstudio, where you can build, govern, and orchestrate all AI agents and agentic workflows at scale.
Agentstudio is a good way to get started with deploying a single agent, and it’s also a good way to explore subsequent steps in realizing the increasing value of non-deterministic systems within the context of traditional deterministic software systems.
And if you use Boomi Meta Hub as your semantic foundation and Boomi Data Hub for your single source of truth, you can ensure the high data quality and unified business context you need for increasingly complex agentic processing tasks.
The Intellyx Take
When talking about non-deterministic AI, people will often say that they want better control, or a way to validate results. But what they really want are solutions that work well for new applications rather than replacing existing applications.
Strategic new use cases typically involve re-thinking and re-evaluating current processes and workflows, moving from rigid and brittle logic to adaptable and autonomous workflows. This includes providing agents with trusted data and deep, contextual business awareness.
Achieving business value and a return on investment for agentic AI requires taking small steps, building on them for complex, autonomous workflows with continuous evaluation and improvement.
But it all has to start with a recognition that non-deterministic AI agents are better suited to some use cases than others so that initial steps successfully lead to more and more steps.
It helps to have a technology partner such as Boomi, who understands the difference and supports your requirements across deterministic and non-deterministic use cases, separately or in combination.
See how Boomi Agentstudio helps you design, deploy, and manage AI agents with built-in governance and control.
Copyright © Intellyx B.V. Boomi is an Intellyx customer. No AI was used to write this content. Image by Google Gemini.