Article · AI and your team

    Does my small business need autonomous AI agents running whole jobs without supervision?

    By Nina Huchthausen, Strategic AI Partner & Trainer · Published · Updated

    "Should I be building AI agents? Everyone keeps saying agents are the future, and I feel like I'm already behind."

    I hear some version of this every week now, usually right after someone has watched a video of an AI that apparently runs an entire business on its own while the owner sits on a beach. If that is you, you are in very good company, and I want to say something that might upset a few AI bros: I don't think the vast majority of small businesses need autonomous AI running whole jobs without anyone watching. Not to start with, and for a lot of jobs, not for a long time.

    What they need is AI that takes the draining, repetitive work off their people, with a person reviewing it at the moments that matter. Here is why, and where to start.

    What actually is an AI agent?

    There is no single, agreed definition, and no legal one. Governments, researchers and the AI companies themselves all describe agents differently. What they broadly share is the idea of AI that can plan and take steps on its own, using your tools and systems, rather than only answering a question when you ask it.

    That is not me being vague. Singapore's media and technology regulator has said plainly, "There is no consensus on what defines an AI agent" [1], and the definitions in proposed laws overseas vary depending on what each one is trying to regulate [1]. Even the companies building the technology describe it in their own way:

    OpenAI (ChatGPT)

    How they describe it

    "Agents are systems that independently accomplish tasks on your behalf." [2]

    Anthropic (Claude)

    How they describe it

    Splits it in two: workflows, "where LLMs and tools are orchestrated through predefined code paths", and agents, "where LLMs dynamically direct their own processes and tool usage" [3]

    IBM

    How they describe it

    "AI assistants are reactive, performing tasks at your request", while agents can keep working towards a goal on their own after you kick them off [4]

    So if someone tells you their product is an "agent", it is a fair question to ask what it actually does on its own, and what it does not.

    Is it an AI agent or an AI assistant, and does the label matter?

    The label matters far less than two questions: how much does the AI do on its own, and where does a person step in? Autonomy is a dial, not a switch. You decide where to set it for each job, and you can turn it up as your confidence grows.

    Researchers at the University of Washington have mapped this as five levels, based on the role the person plays [5]. In plain words:

    1

    The person's role

    Operator

    What it looks like in a small business

    You drive, and AI helps when you ask. Chatting with AI to draft an email.

    2

    The person's role

    Collaborator

    What it looks like in a small business

    You and AI plan and do the work together.

    3

    The person's role

    Adviser

    What it looks like in a small business

    AI leads the work, and you give feedback and direction along the way.

    4

    The person's role

    Approver

    What it looks like in a small business

    AI does the job, and comes to you when it is stuck or needs your sign-off.

    5

    The person's role

    Observer

    What it looks like in a small business

    AI runs the job with no way for you to step in, apart from an emergency stop.

    Their key point is that an AI's level of autonomy is "a deliberate design decision, separate from its capability" [5]. A very capable AI can still be set up to stop and check with you. I have written more about this play in knowing when to step back in.

    That is how I build with my clients. Some of their assistants run on their own every hour or every night, which by some of the definitions above makes them agents. But they are designed to stop at the points that matter and hand over to a person. I call them assistants because their job is to serve your people, not replace them.

    Does my small business need autonomous AI running whole jobs without supervision?

    For most small businesses, no, not to start with. AI can only follow what you teach it, so it is only as good as your process. If a job still lives in someone's head, or you are still working out how it should be done, an unsupervised AI will repeat the gaps at speed. Start with oversight, and earn your way to more autonomy.

    Small businesses are messy and human. People call in sick, customers change their minds, and priorities shift on a Tuesday afternoon. The way you price a tricky job, or the customer who always needs a phone call instead of an email, is often not written down anywhere. If you cannot teach it properly yet, handing it to AI to run on its own is tricky, and that is how you end up with what I call a high-speed hot mess.

    The big end of town is finding this too. Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls [6].

    And Australia's own guidance points the same way:

    • The Australian Cyber Security Centre, with its partner agencies in the US, UK, Canada and New Zealand, advises organisations to "Deploy agentic AI incrementally, beginning with clearly defined low risk tasks", and to "Prevent agents from autonomously executing high impact actions or outputs without prior human approval" [7].
    • The National AI Centre's guidance for businesses says to make sure a person oversees an AI system "in a way that matches how much autonomy the system has" and how high the stakes are, with clear points where people "can pause, override, roll back or shut down AI systems" [8].

    None of that says don't use AI. It says start small, keep people in charge, and build trust one job at a time.

    Where should my business start with AI?

    Start with a task, not a technology. Find the jobs that drain your people, the ones they never get to, the ones they avoid, and the ones where mistakes keep happening. Pick one, hand it to an AI assistant with clear stop points, and let your team review, correct and feed back until you trust it.

    Those four kinds of task are where the time and the stress are hiding:

    • Draining work, the repetitive, manual jobs that eat the best hours of your day.
    • Work nobody gets to, the follow-ups and checks that keep sliding to next week.
    • Work people avoid, the dreaded job that sits on someone's shoulders all week.
    • Work where mistakes happen, usually because it depends on someone remembering.

    Here is what that looks like in real businesses.

    Aus Patios Direct, lead replies. Every new enquiry used to take Dave about 40 minutes to draw, price and reply to. Now an assistant runs every hour on its own, prices each lead from Dave's own quoting calculator and drafts a reply in his voice. Then it stops. It never sends anything to a customer. Dave checks and sends every one himself. In his words: "Yeah, absolutely broke ground that thing because that would have taken me 40 minutes to do a lead, and now it's sitting here and it's like done in instantly." In its first week it also caught a customer who had given their dimensions backwards, and turned up a bug in Dave's calculator so he could fix it.

    R&A Marine Services, the daily site report. Martin used to spend up to two hours a night turning the day's WhatsApp updates into a site report, sometimes in the early hours of the morning. Now an assistant builds the report in his template every evening, shows who reported every figure, and flags gaps rather than guessing. Martin checks and approves it in about 15 minutes. "Now it is giving me what I want. This is the format that I was wanting Claude to continue using."

    What does a good stop point look like?

    A stop point, or stopgate, is a planned moment where the AI pauses and a person reviews, approves or corrects its work before anything goes further. It is where your judgement stays in the job. It catches mistakes early, and every correction makes the AI better at the next one.

    Good stop points sit in a few predictable places:

    1. Before anything leaves the business, such as an email to a customer, a quote or a payment.
    2. When something is missing or doesn't match, so the AI flags it instead of guessing.
    3. Before anything that is hard to undo, such as deleting records or changing money.
    4. At the end of each run, as a short summary a person can glance over.

    At Change for Life, a nightly assistant checks that every session has an invoice and every payment has landed in the bank. The daily reconciliation went from about an hour to 15 minutes. The first time it ran, Malini told me: "he was amazed by the reconciliation.. he checked it all.. it was correct.." That checking is exactly what built the trust. And when something is missing or a system is disconnected, the assistant does not guess. It creates a task and sends an alert, so a person decides what happens next.

    The same business has an assistant sorting Aaron's inbox into a ranked briefing three times a day, with replies drafted in his own voice for him to approve. His inbox time dropped from two to three hours a day to about an hour. Malini put it best: "But you don't know the best part of it. He didn't tell you. I have not reminded him about anything once."

    That is what good AI looks like in a small business. Not something that replaces the people, but something that means they finally get to stop carrying everything.

    Humans on top. AI underneath.

    When can I let AI do more on its own?

    When the process is clearly defined, the AI has done the job right with a person checking for weeks, and the step you are handing over is low risk and easy to undo. Then turn the dial up one step at a time. Keep a person approving anything high impact or hard to reverse.

    This is the order I walk every client through:

    1. Pick one task that drains your people. (If you're not sure where to look, start with how to find the work you love and hand over the rest.)
    2. Map how it actually gets done today. Not how it should be done, how it actually is. This is where you find the bits that only live in someone's head, and it is why your processes matter before you bring in AI.
    3. Train an assistant to do the first pass. Not chat with it. Train it on your process, your voice and your rules, the same way you would onboard a new team member.
    4. Put in your stop points, and review everything at first. Correct it as you go, and feed those corrections back into its training.
    5. Hand over one more step only when it has earned it. Something small and reversible, like logging the quote date after you send. Then watch it.

    Start small. Get the process right. Build something useful. Then build on it.

    You aren't behind. The groundwork you do now with one well-trained assistant and clear stop points is exactly what makes more autonomy safe later.

    And if someone promises you an AI agent that will make you $10 million while you sit on a beach doing absolutely nothing, I'd be asking a few bloody questions.

    I'm all for AI that makes businesses run better. I'm just far more interested in AI that works in the real world, on a messy Tuesday afternoon, than in something that looks impressive in a demo and falls apart the minute your business does something unexpected.

    More magic. Less mess.

    Not sure which task to hand over first?

    My free AI Opportunity Map works a little like an AI interviewer. It walks through your business with you and shows where AI could safely make the biggest difference for you, and how much time you could get back.

    Get it here 👉 Your AI Opportunity Map

    Would you rather talk it through? Let's have a chat. No prep. Just an honest conversation.

    Book a free 30-minute call

    Keep reading

    Nina Huchthausen, Strategic AI Partner & Trainer at Set to Growth

    Nina Huchthausen

    Strategic AI Partner & Trainer, Set to Growth

    Nina is a process engineer by trade. For 16 years she has helped businesses streamline their processes and systems, and now she trains AI to take the manual, repetitive work off their team's plate. Her clients feel the draining work lift, and get either the room to grow and scale, or the same revenue in a 7 hour day instead of 12.