Industrial organizations are under pressure from three directions at once: persistent labor shortages, the need to digitize more work, and the demand to increase output with limited resources. Artificial intelligence appears to offer an answer, but many leaders are still trying to separate practical applications from marketing noise.
In Episode 28 of MetaPod, host Ron Crabtree speaks with Scott McIsaac and Len Landale, co-founders of Helios Core, about how AI agents can perform real work inside a business. Drawing on their experience in IT managed services, the guests explain how organizations can move beyond chatbots toward intelligent automation that interprets inputs, makes decisions, interacts with systems, and completes workflows.
Their central recommendation is straightforward: think strategically, begin tactically, and do not automate work that the organization cannot clearly explain.
When leaders are told to adopt AI, a common response is to purchase a general-purpose tool for employees and call that the strategy. McIsaac argues that simply buying an AI assistant does not create productivity on its own. Employees need training, practical guidance, approved tools, and a clear understanding of what the technology can and cannot do.
AI adoption also cannot belong exclusively to the IT department. The useful opportunities sit inside business processes, so the people who understand those processes must help choose the problems, define the outcomes, and identify the information the technology needs. A cross-functional group can bring together operational, technical, security, and business perspectives without turning the effort into an endless planning exercise.
The strategic question is where AI can improve a meaningful business result. The tactical question is which narrow use case provides a safe place to begin.
Landale recommends looking first for work that arrives in volume, follows a recognizable pattern, and requires relatively little judgment. These tasks consume capacity but do not always make the best use of experienced employees. Examples discussed in the episode include password resets, routine ticket processing, pricing activities, overnight monitoring, and work that expands sharply during seasonal peaks.
This is the crawl-walk-run approach. An organization can deploy a focused tool or agent, measure whether it works, and then broaden its scope. Attempting to automate an entire end-to-end process at the beginning is possible, but it is more expensive and depends on systems, data, and process documentation that may not yet be ready.
Starting small can also reduce employee anxiety. When AI removes repetitive processing, triage, and information gathering, people can spend more time on exceptions, decisions, customer needs, and other work requiring experience.
The episode distinguishes an AI assistant from an AI agent. With an assistant, a person supplies information, asks questions, interprets the response, and decides what to do next. An agent is given the knowledge, permissions, and tools needed to carry a process forward with a greater degree of independence.
Landale illustrates this with an IT service request. An agent could read a password-reset ticket, verify the requester through an approved method, call the appropriate tool to unlock the account, confirm that the action worked, and document the result. McIsaac describes agentic AI more simply as intelligent automation: automation that can reason through less structured inputs and select a suitable path instead of following only one rigid sequence.
That independence creates value, but it also increases the importance of clear processes, controlled system access, testing, and security.
Many organizations rely on tribal knowledge. An experienced employee may know which material to select, how to price an unusual request, or what exception applies in a particular situation, even though the rule has never been documented. That arrangement becomes fragile when the employee retires or when the organization tries to automate the process.
Historical emails, completed quotes, and prior work can help preserve expertise, but the organization still needs to make its specific decision rules explicit. An AI system cannot reliably follow an assumption that exists only in someone’s head. Without grounding, it may guess or produce an incorrect result.
Process legibility therefore means documenting the work from beginning to end: the inputs, steps, decisions, system interactions, exceptions, and edge cases. Existing documentation should be examined with a deliberately literal eye. An instruction that seems obvious to an experienced employee may be ambiguous to a system.
Once the process has been defined, teams can run test cases at scale, expose unclear instructions and unusual conditions, refine the workflow, and move gradually toward productive use.
The value of automation is larger than a simple labor calculation. Leaders should ask what a process produces, what it costs to execute, and what happens when it is late, incomplete, or wrong.
The episode uses employee offboarding to demonstrate the difference. Removing access may require coordinated actions across email, licenses, and business systems. If one step is missed, the organization may retain an unnecessary license or, more seriously, leave a former employee with access. The cost of the outcome includes risk, error, delay, and exposure—not just the time required to perform the task.
The same reasoning applies to service capacity and downtime. An agent that performs early triage, reviews available information, and prepares a recommended action can reduce the time an on-call employee spends diagnosing a problem. In a manufacturing setting, telemetry could help identify why equipment stopped and give a technician better information before arriving at the site.
Useful measures may therefore include throughput, service speed, downtime, error reduction, security exposure, employee workload, and the consequences of process failure.
AI tools do not automatically make employees more capable. People need enough training to understand how to use the tools, evaluate their output, and recognize when information or instructions are missing. Training is one of the basic conditions for adoption—not an optional activity after deployment.
McIsaac and Landale present AI primarily as a way to increase capacity. Agents can absorb repetitive tasks, prepare diagnostic information, and handle part of an overnight or seasonal workload. Employees remain responsible for the higher-judgment work and can intervene when a decision or exception requires human expertise.
That model changes the conversation from replacing people to improving the mix of work they perform.
Security cannot be added after an AI initiative is already operating. McIsaac advises organizations to use properly configured business or enterprise accounts, understand where their information goes, and establish policies and processes that give employees an approved way to use AI.
A blanket prohibition can create a different problem: employees may work around the policy and use unauthorized tools. A stronger approach combines accessible approved technology with training, configuration, data controls, and clear expectations.
Security should be considered alongside every use case, especially when an agent can access systems or take actions. The more independently the technology operates, the more important it becomes to define its boundaries and verify how it handles information.
An unsuccessful pilot does not necessarily mean that AI has no value. The guests note that projects can disappoint because the organization attempted too much, selected the wrong starting point, lacked adequate documentation, or failed to prepare the process and its data.
Organizations should continue identifying focused use cases, testing them, and learning from the results. They should also consider experienced outside support when internal teams cannot keep pace with the technology while running the core business.
The competitive pressure is not limited to today’s rivals. Landale warns that new businesses may be designed around AI from the beginning, without the same legacy systems, accumulated system debt, or concentration of tribal knowledge. Waiting carries its own risk.
The episode’s advice can be reduced to a disciplined sequence: bring the right business and technical stakeholders together, identify high-volume work with limited judgment, document how that work is actually performed, define the expected outcome, train employees, build in security, test thoroughly, and expand only after demonstrating value.
AI agents can do real work, but useful automation begins with work the organization understands. The companies most likely to benefit will be those that combine ambition with process clarity, measurable outcomes, employee preparation, and appropriate control.
To learn more about Scott McIsaac, Len Landale, and Helios Core, visit helios-core.com or contact the team at info@helios-core.com.
MetaPod focuses on the recurring challenges facing industrial organizations: labor shortages, the digitization of work, and the pressure to do more with less. Follow MetaPod for more conversations about operational improvement, AI application, and organizational transformation.
With one phone call, email, or clicking a link schedule a time to speak with the MetaExperts® sourcing expert. You are just minutes away from getting your improvement initiative started or re-energized.
MetaExperts® is a service brand of MetaOps, Inc.
©2026 MetaOps, Inc. All rights reserved.
