Who’s Managing the AI Agents Your Company Pays For?
Written by: Jules Flesner
AI agents vs. agentic AI: The terms sound nearly identical, and because the concepts are still relatively new, their definitions vary across industries and technology companies.
They describe technology already becoming standard within the software and apps businesses have used for years, with the ability to access data, communicate, make decisions, and take action.
Every leader responsible for protecting an organization now needs to understand what these terms mean, along with their benefits, risks, ethical concerns, and true costs.
Website chatbots common in the 2000s and early 2010s typically followed prewritten rules and scripts. They could answer expected questions but were easily stumped by unfamiliar wording or requests, defaulting to a human handoff.
Modern AI runs on fundamentally different technology. It can interpret everyday language, recognize patterns, work through requests, and respond conversationally. When built into a company’s business software and given the proper permissions, it can also take action.
That is where AI agents begin.
What is an AI agent?
Think of an AI agent as a software worker that can understand a goal, make decisions, communicate in natural language through text or voice, and take action.
An AI agent, or “digital worker,” has a specific job. It may interact directly with customers or work exclusively behind the scenes to assist employees.
A basic chatbot from the 2000s might have simply informed a customer how to request a refund. Today, an AI agent can verify eligibility, process the refund, update the customer’s account, and send confirmation.
Like any worker, an AI agent needs clear instructions, firm boundaries, reliable data to work from, and proper oversight. (Human) specialists have to define its role, system access, spending limits, compliance requirements, approval rules, and when it must stop and hand off the matter to a qualified person.
As of today, customers may communicate with an AI agent by phone call, email, text, or online chat without realizing they’re talking to software instead of a person. U.S. disclosure requirements currently vary by state, industry, audience, and use case. Even where no specific disclosure rule applies, allowing software to misrepresent itself as a person can create consumer-protection, compliance, and reputational risk.
The true cost of an AI agent includes more than its software subscription.
It also includes implementation, integration, data cleanup, security controls, testing, training, compliance review, ongoing monitoring, and the human expertise required when the system reaches its limits.
What is agentic AI?
Agentic AI describes the broader capability of AI systems to plan, make decisions, use tools, adapt to changing information, and complete multiple connected steps with limited human direction.
Here’s an example from the rental housing industry:
An agentic AI software system can receive a resident’s maintenance request, communicate with the resident to assess urgency, check unit and warranty records, schedule the appropriate technician, order approved parts, provide updates, and immediately alert the designated contacts/specialists if it detects an emergency. It can also read vendor invoices, assign expenses to the correct property, unit, and work order, route bills for approval, and flag charges that exceed approved limits.
An example in the retail industry:
A skincare company selling products online might use several specialized AI agents working together. A shopping agent recommends products based on the customer’s stated skin concerns, an inventory agent checks availability, a payment agent processes the purchase, a fulfillment agent coordinates shipping, and a customer-service agent handles delivery updates, returns, or refunds. Human employees review allergy concerns, adverse reactions, unusual refund requests, and any situation requiring medical or professional judgment.
Agents operate inside a larger system of data, software, permissions, policies, and human approvals. Together, these agents, systems, permissions, policies, data sources, and human controls form an agentic AI system.
Should AI replace human workers?
Agentic AI raises a serious ethical and operational question worth discussing! Some jobs are already changing, and others are disappearing.
The detail largely missing from today’s commentary is that many organizations entered the agentic AI era already understaffed. Leadership had previously reduced, eliminated, or never approved the positions needed to manage the workload. In those cases, saying that agentic AI is “replacing a human team” misses an important point: the fully staffed team never existed!
Some companies also overhired during the late 2010s business boom, creating impressive-sounding titles with little authority or lasting purpose.
The more common situation I see is employees who were already working in skeleton-crew departments, or local small business owners winging it on their own, stuck moving business tasks between poorly designed, disconnected systems.
That’s why agentic AI feels so appealing. In theory, it offers a cheaper fix: absorbing unfinished work, relieving overloaded employees, and making the earlier decision not to hire more people seem justified.
It also raises a critical question:
If (human) workers already struggle with broken systems and an inadequate training environment, how is a software worker supposed to succeed?
Who still has to fix missing or unreliable data, take responsibility for poor software onboarding and training, initiate and resolve difficult conversations between departments, track constant compliance changes, push software vendors to correct bugs and bad system design, show up to defend the company’s decisions in court when a system fails?
Humans!
Human expertise should never be treated as disposable simply because a software vendors’ marketing claims that their (still-evolving, still-being-built) AI agents can perform the work.
Learning the terminology is only the starting point.
Leaders need to understand where agentic AI is already affecting their business and prepare to manage its expanding role in the software they use for daily operations.
Which leads me to…
Five leadership patterns creating unnecessary risk
No owner or organizational leader needs to become AI engineer overnight!
They do need to know which agentic systems and apps are already in place inside their organizations, what data those systems can access, which decisions they can influence, and who remains accountable for the outcome.
After a year of systems design consulting, these are the top five leadership patterns I have seen across private companies and nonprofit organizations (so, if any sound familiar, you’re far from alone):
Stuck 20 years in the past. They hear “tech consultant” and think I’m the same specialist that fixes their office printer or check scanner. They hear “marketing” and expect a $500 surface-level website update solves everything.
Using tech they don’t understand. The software providers they’re set up with have already enabled AI features that take action, communicate with customers, or affect company policy, terms, and disclaimers before company leaders fully understand what the features do. Leaders have no budgeted plan to track these ongoing updates, audit risks, test the new feature, set boundaries, train employees, or maintain expert oversight.
Using a fraction of the software subscriptions they’re paying for. They approved or upgraded to agentic AI or other online tools, but have no practical plan to fully implement or maintain them.
Believing miracle sales rep claims. Some later discover they were already paying for similar AI or automation features through another software contract.
Assuming everyone else has it figured out. Leaders who admit they are tech-challenged assume their more tech-savvy employees and new hires will somehow arrive telepathically trained on dozens of subscription-specific, custom-configured, constantly changing software platforms packed with new AI features, never setting aside the several hours each week or month that it actually takes for training and testing. Meanwhile, employees repeatedly ask for training and support while their valuable operational feedback disappears into a black hole, year after year.
Mistakes are inevitable. No company gets everything right. The real failure is continuing these behaviors from a position of authority and refusing to seek qualified systems help.
Did You Learn The Difference?
An AI agent is a software worker that can understand requests, make decisions, communicate, and take action. Agentic AI is the term for the broader system that enables one or more AI agents to plan and coordinate multiple actions toward a goal. (These two terms will probably be used inconsistently for some time.)
Do you plan to be in business five years from now?
If the answer is ‘yes,’ are you preparing for where agentic AI technology will take your company and your competitors?
Agentic AI is already here. Ask yourself:
Have our job descriptions and training, including my own, caught up with the unprecedented AI changes already happening inside the business?
Do I have the right people reviewing our software subscriptions and overseeing every app the business depends on?
Could anyone clearly explain which software services can access our company data, website content and visitor data, or employees’ personal devices?
Are customers receiving conflicting information about our organization or services from employees, online listings, AI tools, our website, or other marketing materials?
Would a regulator, attorney, investor, or board member find our records clear and defensible?
What budget can I approve today to bring in a specialist who can uncover and fix hidden problems before (fingers crossed) they become devastatingly expensive operational, financial, security, or compliance failures?
Organizational leaders who answer these questions honestly, and engage an AI systems specialist proactively, are better prepared to protect their companies, redesign work thoughtfully, evaluate vendors, and keep a competitive advantage for years to come.
Is your organization ready for agentic AI?
I help private companies and nonprofits audit their software subscriptions and AI tools, workflows, employee training, and human oversight before 'invisible issues’ become expensive failures. Request an initial consultation for your organization.
