The 3 symptoms of an agentic process (and when to stick with standard automation)

When you’ve been immersed in new ideas, whether you're spent days at a conference, finished a deep-dive YouTube rabbit hole, or walked away from a conversation that finally clicked, inspiration makes your brain feel excited about the potential of what’s next. But it’s hard to get started with what’s next when you still have daily stand-ups, full email inboxes, and a backlog that didn't pause while you were thinking big.

I know this feeling, and I also know that making progress requires a structured plan to move ideas into action. For those who attended Imagine, I wanted to make sure that they left Dallas understanding the power and potential of AI agents, and how to identify also which processes actually require an AI Agent. So, that’s what I spoke about.

I believe that if you can learn to identify the symptoms of an agentic process, you can accurately diagnose opportunities and prescribe exactly where or whether an AI agent belongs in a given workflow.

The diagnostic isn't complicated, but it does require asking the right questions in the right order. Let’s take a look at the primary symptoms of an agentic process first.

Symptom 1: Ambiguous or Variable Input

Traditional robotic process automation (RPA) thrives on rules-based predictability. These types of automations have proven to be extremely valuable for business processes where the inputs are strictly controlled and formatted. However, when a business process or user deviates from this formatting, the automation may be more prone to errors.

Agentic processes introduce the ability to interpret ambiguous or variable inputs. This is especially important for situations where you can’t predefine or templatize exactly what is coming into the system. This requires the interpretation of tone, sentiment, and nuance in addition to understanding the intent of the request. AI agents handle this variability naturally, making them the optimal solution for customer support emails, non-standard vendor contracts, and IT help desk tickets where incoming requests may contain multiple, intertwined issues or requests.

The diagnostic question you need to ask: Do I know what is coming in and how it is formatted before it arrives? If the answer is no, you need an agent.

Symptom 2: Contextual Accumulation

Most processes that are automated require a fully predefined map: I do this, then this, then that. Before a single step executes, developers create a complete path accounting for every possible variable, conditional branch, and edge case.

This structured logic is incredibly effective for linear tasks. But when the business process requires more dynamic thinking, automations would have to push the work back to a human operator. The challenge is that complex work rarely follows a straight line, and what you learn in one step, may change how subsequent steps should be executed.

Agentic processes dynamically adapt their execution path based on information discovered mid-flight. What the system learns from each tool execution and at each step determines the subsequent actions an agent needs to take. Human workers and AI agents handle contextual accumulation seamlessly by building a working memory of the transaction.

For example, when a user reports that they are locked out of an IT system, the underlying cause varies. The answer isn’t always “Run an automation to unlock them”. The user might have forgotten their password, an underlying system may be down, the system might have detected a security threat or maybe the user never actually had access provisioned to that app in the first place. An AI agent has the ability to think on its feet - investigating the root cause, and that unique discovery dictates the exact remediation path required for that specific ticket.

The diagnostic question you need to ask: Does what I discover mid-process change what I need to do next? If the answer is yes, you need an agent.

Symptom 3: Multi-Source Synthesis

Automations that execute integrations between systems are excellent at moving data. They pull a value from one system and push it to another efficiently and reliably. What they can't always do well is understand how the data in one platform changes the meaning of data in another.

AI agents synthesize context from multiple independent systems to make informed judgment calls. This third symptom occurs when no single data repository provides enough information to resolve the transaction independently. The correct answer only emerges when data from various platforms is brought together and evaluated in context.

For Accounts Payable invoice processing, matching an invoice against a purchase order is insufficient. An agent synthesizes data from the delivery manifest, the vendor contract, and the inspection report to confidently approve the payment. Similarly, a loan evaluation requires synthesizing credit scores alongside assets-in-management and work history to form a complete picture.

The diagnostic question you need to ask: Could one system or one person alone make this call? If the answer is no, you need an agent.

The Grey Area: Evaluating Secondary Symptoms

Not every automation pipeline opportunity is black and white. If your business process lacks a primary symptom, it does not immediately rule out an AI agent. However, if your process exhibits one of the following secondary symptoms, it may mean you need to do some further evaluation. You may need an agent, but you must evaluate the cost-benefit analysis:

• Secondary Symptom 1: Exception-Heavy Processes: Does more than 30% of the workflow require human judgment? When exceptions become the workload, such as procurement workflows constantly disrupted by emergency sole-source justifications, the process has outgrown standard conditional logic.
• Secondary Symptom 2: Volume Exceeds Capacity: Are you settling for sampling when you should be covering 100%? For instance, financial firms required to supervise advisor communications often review only 5-10% of correspondence manually. Agents enable full coverage without headcount scaling. You might be thinking, “great, now we can do more compliance related work.” Sure, that’s an option, but what if that correspondence was also a way for you to identify training opportunities for your staff or detect churn/upsell opportunities? Now the benefits of 100% review start to align to meaningful commercial outcomes.
• Secondary Symptom 3: Complex or Evolving Logic: Have you had to rebuild your automation logic more than once in the last 12 months? If rules change frequently (e.g., trade compliance teams updating tariff codes 20+ times a year), maintaining standard automation becomes too costly. Agents reason against policy documents instead of hard-coded logic and rules. This gives you the flexibility to quickly keep your processes up to date without having to re-work all of your logic whenever there are policy or procedure updates.

The Monday Morning Playbook

A business process absent of these primary or secondary symptoms may not require an AI agent. That’s perfectly fine. Intelligent automation (automations leveraging AI for a discrete purpose in a deterministic automation) remains a highly effective solution for rigid, rules-based workflows.

The goal is to evaluate the technical requirements of your business process and apply the right technology to the right problem at the right point in your workflow.

To move from excitement about a new idea and inspiration to programmatic momentum, you need a structured approach to evaluate your backlog. During my keynote at Imagine Dallas this May, I shared the "Monday Morning Playbook," a comprehensive guide detailing these primary symptoms, additional secondary symptoms, and an agentic impact measurement framework.
 

When you review your opportunity pipeline this week, follow these three steps:

• Step 1. Pick a few processes: Select active items from your backlog, something new that just came in, or a workflow you are currently building.
• Step 2. Run the diagnostic: Use the playbook to ask yourself the three symptom questions and determine exactly where an AI agent might (or might not) fit within the process.
• Step 3. Prioritize on impact: Use the measurement framework to evaluate the multi-dimensional benefits, ensuring the opportunities you pursue align directly with your executive team’s key objectives.

By running your pipeline through this diagnostic framework, you ensure your automation program focuses on building solutions that drive true strategic momentum.