By Chuck Gallagher — Business Ethics Keynote Speaker and Trainer
TL;DR: Chuck Gallagher, AI ethics speaker and author, argues that when small businesses let AI agents take actions on their behalf, the accountability never transfers to the software, so every agent needs defined permissions, approval limits, activity records, and a named human owner.
Picture a family-owned heating and air company with nine employees. The owner, worn thin after a brutal summer, sets up an AI agent to handle customer follow-ups. It can read the inbox, check the schedule, book appointments, and send replies. For three weeks it works beautifully. Fewer missed calls. Faster responses. Happy customers.
Then a customer writes in, angry about a repair that didn’t hold. The agent, built to resolve complaints quickly, replies that the company will waive the service fee, replace the part at no charge, and have a technician out by noon tomorrow. Nobody approved any of that. The noon slot was already booked. And the customer saved the email.
Now the owner has a choice. Honor a promise he never made, or break a promise his business did make. Either way, somebody pays.
As an AI ethics speaker and author, I think that scenario is where a lot of small businesses are headed, whether they realize it or not. Atlassian’s State of Product 2027 report, built on a Wakefield Research survey of 1,000 product professionals in the United States, Germany, and France, found that 52% of teams using AI now describe it as a task runner, meaning a tool that completes multistep tasks and takes approved actions in third-party tools. Another 13% call it an autonomous agent that pursues goals with minimal instruction or supervision. Only 7% still see it as a basic assistant that drafts emails and answers questions.
What Changes When AI Stops Drafting and Starts Doing?
A draft is a suggestion. A human reads it, judges it, and decides whether it goes anywhere. An action is different. When an agent sends the email, books the appointment, issues the refund, or updates the customer record, there’s no second look. The decision has already been made. That’s the shift. AI moved from the passenger seat to the driver’s seat, and nobody changed the rules of the road.
The same report shows that supervision hasn’t gone away. Asked what they’ve had to do with the time AI saved them, 49% said they constantly check in on AI agents to make sure they stay on task, and 48% said they retrace an agent’s steps to figure out what went wrong. Roughly half of these professionals are spending their reclaimed hours babysitting the very tools that were supposed to free them up.
Why Should Small Business Owners Pay Close Attention to AI Agents?
Those survey respondents work at companies with 500 to 10,000 employees, where IT staff, compliance teams, and layers of managers can share the watching. A small business owner usually has none of that. When an owner hands scheduling, customer follow-up, or billing reminders to an agent, the temptation is to treat it like a dishwasher. Load it. Close the door. Walk away. But an agent that talks to your customers isn’t an appliance. It speaks for you.
Who Is Accountable When an AI Agent Makes a Promise to a Customer?
Let me be clear. Delegating a task does not delegate the accountability. The agent didn’t hire itself. Somebody chose the tool, granted its access, and decided what it could say and do. That person, and the business behind them, owns the outcome.
The legal system has already said as much. In Moffatt v. Air Canada, decided in February 2024, British Columbia’s Civil Resolution Tribunal heard the airline argue that its website chatbot, which had given a grieving customer wrong information about bereavement fares, was effectively a separate entity responsible for its own actions. The tribunal rejected that argument, held the airline responsible for all the information on its website, and ordered it to compensate the customer. That was a chatbot giving information. An agent taking actions raises the stakes.
So go back to the heating and air company. Who approved the agent’s authority to waive fees? Nobody did, explicitly, and that’s the problem. The owner handed it the keys to the inbox and the calendar but never drew a line around what it could promise. The harm belongs to the business, and the repair falls to a human, someone who has to call that customer, own the mistake, and decide what the company can honorably do. Trust me, the customer doesn’t care that it was software. They care that your company gave its word.
How Should a Small Business Set Boundaries for an AI Agent?
The answer isn’t to avoid agents. It’s to supervise them the way you’d supervise a new hire. Start with permissions. Write down exactly what the agent can touch and what it can’t, such as booking appointments but never changing prices, or sending reminders but never offering discounts. Give it the least access the job requires.
Next, set approval thresholds. Any action involving money, a refund, a contract term, or a commitment beyond your standard policy should stop and wait for a human yes. Then keep activity records. Every message sent and every action taken should be logged somewhere a person can read it, because when something goes wrong you’ll need to retrace the steps, and nearly half the professionals in Atlassian’s survey are already doing exactly that.
Finally, and most important, name a human owner. Not “the team.” Not “IT.” One person whose name is attached to that agent, who reviews its work, answers for its mistakes, and has the authority to shut it off.
As an AI ethics speaker and author, I’ll put it plainly. Every choice has a consequence, and choosing to let a machine act on your behalf is still your choice. Your agent can work around the clock. It can’t carry the weight of your word. That part stays with you.
Frequently Asked Questions
What is an AI agent for small business?
An AI agent is software that goes beyond drafting text to taking actions, such as sending emails, booking appointments, or updating records in other tools. In Atlassian’s State of Product 2027 report, 52% of product teams using AI described it as a task runner that completes multistep tasks and takes approved actions in third-party tools. That shift from suggesting to doing changes how much oversight a business needs.
Who is liable when an AI agent makes a mistake with a customer?
In general, the business that deploys an agent is responsible for what it says and does on the company’s behalf. In Moffatt v. Air Canada (2024), a Canadian tribunal rejected the airline’s argument that its chatbot was a separate entity and held the airline responsible for the misinformation. Owners should assume their agents speak for them and consult an attorney about their specific situation.
How much supervision do AI agents need?
More than many owners expect. In Atlassian’s survey of 1,000 product professionals, 49% said they constantly check in on AI agents to keep them on task, and 48% said they retrace agents’ steps to diagnose problems. Agents save time, but they don’t remove the need for a human to watch the work.
How do you set permissions and approval limits for an AI agent?
Define in writing what the agent can access and what it can do, granting only the access the task requires. Require human approval for anything involving money, refunds, contract terms, or commitments outside standard policy. Keep a readable log of every action so problems can be traced and corrected.
Why should every AI agent have a named human owner?
A named owner ensures someone reviews the agent’s work, answers for its mistakes, and has the authority to shut it down. Chuck Gallagher, an AI ethics speaker and author, argues that delegating a task to AI never delegates the accountability for it. When no one owns the agent, no one owns the harm it causes.
Bring This Conversation to Your Organization
If your business is handing real work to AI agents, the time to define their limits is before the first customer complaint, not after. I help leadership teams, associations, and conference audiences build accountability into AI adoption, from who approves an agent’s authority to who answers when it gets something wrong. To bring a keynote or workshop on AI ethics and accountability to your organization, connect with me, Chuck Gallagher, at ChuckGallagher.com.
Research sources: Atlassian, The State of Product 2027, based on a survey of 1,000 product professionals conducted by Wakefield Research, August 13–26, 2026; Moffatt v. Air Canada, 2024 BCCRT 149.
Five Questions for Reflection
- Which tasks in your business are AI tools already performing, and which of those involve acting rather than just suggesting?
- If an AI agent made a promise to one of your customers today, who in your organization would find out first, and how?
- What specific commitments, such as discounts, refunds, or deadlines, should never be made without a human’s approval?
- Who is the named owner of each AI tool your team uses, and do they know it?
- Is it fair to hold a business to a promise its software made without anyone’s approval? Why or why not?
