Funal← Journal
AIAI AgentsService BusinessAutomation

Can an AI actually run your business?

JUL 19, 2026·10 min read·Alex Boquist

The honest answer to 'can AI run my business,' from a founder building the thing. No, it cannot, and the vendors who say otherwise are selling a chatbot in a costume. But it can carry a specific and large half of the job, and here is exactly which half.

No, an AI cannot run your business, and anyone selling you one that does is selling a chatbot in a costume. But it can carry a specific and large half of the job, and being clear about which half is the difference between real help and an expensive disappointment. What a tool like Funal actually does is take the holding, not the judgment. That sentence is the whole article. The rest is me earning it.

I build this software, so read my conclusions the way you would read anyone's who has something to sell: as an argument, not a verdict. But I have a reason to be precise rather than hype you. The businesses that buy the fantasy churn in ninety days, and I would rather you buy the true thing and keep it.

Attention has two parts, and we always confused them

Think about what it takes to serve a client well over months or years. Some of it is keeping them in mind: noticing that the usually chatty one went quiet, remembering that the review is due next week, recalling that last time they mentioned a spouse who handles the decisions, drafting the next message before the thread goes cold, never letting a single relationship drop off the edge of your memory. Call that the holding.

And some of it is the part only you can do: the judgment about what this particular person actually needs, the warmth in the room, the hard call you have earned the right to make, the read on a situation that no record contains. Call that the gift.

Here is the thing we never noticed. These two jobs always got done by the same person, so we assumed they were one job. They are not. The holding is patient, tireless, relentless attention to detail. The gift is judgment and care. The first can be handed to software that never sleeps and never forgets a name. The second cannot, and if you find yourself wanting to hand it off, that is a different problem than software solves. The reason "can an AI run my business" is the wrong question is that it treats one blob called "running it" as a single thing to keep or give away. Split it in two and the real question appears: can an AI do the holding so you are free for the gift. That one has a yes.

Why most "AI employees" fall on their face

Most of the AI staff being sold right now fail, and they fail for a boring reason. They have no idea what is actually going on in your business.

An AI is only as good as what it reasons over. Point it at a typical CRM and you have pointed it at a graveyard of half-filled fields that a busy human updated when they remembered, which was rarely. Business contact data goes stale at about 2.1 percent a month, roughly 22.5 percent a year, according to research HubSpot publishes from MarketingSherpa (HubSpot). So within a year a quarter of what the AI is reading is already wrong, and it has no way to know which quarter. Feed a confident machine stale testimony and you do not get a helpful colleague. You get garbage in, confident garbage out, delivered in a professional tone to a client who trusted you.

This is not a hypothetical. When Carnegie Mellon built a simulated company and turned the best available agents loose on ordinary multi-step office work, the strongest model finished about 30 percent of the tasks on its own and flubbed the rest, a failure rate near 70 percent (The Register, on the university's TheAgentCompany benchmark). Gartner expects more than 40 percent of agentic AI projects to be scrapped by the end of 2027, and names the culprit plainly: hype, unclear value, and a lot of "agent washing," old chatbots wearing the word agent like a badge (Gartner). The number is not surprising once you see the mechanism. An AI working off what humans typed inherits every gap and every lie in the typing.

The ones that work fix the input, not just the model. They derive the record from what actually happened: the calls, the emails, the bookings, the payments. Those events do not depend on anyone remembering to log them. This is the premise Funal is built on, so the AI reasons over reality instead of over testimony. When the record is true because the system keeps it true, the AI stops confidently doing the wrong thing, because it is no longer reading fiction.

"AI staff, not a chatbot" is a claim about jobs

When I say AI staff, I do not mean a question-and-answer box parked in the corner of your screen waiting for you to prompt it. A box like that does nothing until you think to ask, which means it does nothing on the day you are too slammed to ask, which is the day you needed it.

I mean workers. Named, with a specific job, accountable to a number the way a person on your team is accountable to a number. One watches for the client who went quiet and drafts the message that reopens the thread, grounded in that client's actual history. One prepares the recap after the session. One flags the account that is drifting toward the exit before it is gone. And here is the part that separates staff from a toy: it brings the work to you. The follow-up arrives written, not as a reminder that says "follow up with Sarah" but as the actual note to Sarah, waiting for your yes. This is where Funal sits: AI teammates that do the holding and hand the gift back to you. Anything consequential passes through an approval gate. Everything the AI does is logged, attributed, and inspectable, so you are never wondering what it did in your name.

The limits, said out loud

This next part is the part that earns the rest, so I am going to be blunt about it.

It will not replace your judgment. It will not supply your warmth. A vendor who tells you it will is either confused about their own product or lying to close you, and you should treat the claim as the tell it is. The AI does the holding. You bring the gift. That division is not a limitation I am apologizing for. It is the design.

So anything that leaves the building waits for a human. External emails, commitments, promises, anything hard to walk back: those are yours to approve, by default, every time, until you decide otherwise. And you decide. You set how much rope the AI gets, per action and per client, and you can pull it back the moment it does something you did not like. You stay in charge of the autonomy. The software earns more of it slowly, the way a good hire does, or it does not earn it at all.

What to actually demand

If you are evaluating anything sold as "AI for my business," you do not need to understand the model. You need four answers, and a vendor who cannot give them straight is not ready to be near your clients.

Where does it get its truth: typed in by humans, or derived from the events that actually happened. What can it do on its own, without asking you first. What requires your approval before it goes out. And can you see, after the fact, everything it did and why. Get those four answers in plain language. If the vendor dodges, hand-waves, or answers a different question than the one you asked, walk. The dodge is the answer.

The goal was never an AI that runs your business

The goal was never an AI that runs your business. You started this to do the work you are good at, and somewhere along the way the business filled up with a second job made of upkeep and follow-up and remembering, the parts you should never have had to carry by hand. That is the half to give away. An AI that does the holding gives you back the room to run the business, which was always the only part that needed you.

Frequently asked questions

What is an AI agent for business, really?

An AI agent is software given a specific job and the ability to take steps toward it on its own, rather than only answering when asked. The useful ones in a service business are narrow: watch for clients going quiet, draft the follow-up, prep you before a call, flag an account at risk. The honest test is not how smart it sounds but two things: does it reason over what actually happened or over what someone typed, and does it bring consequential work to a human before acting. An agent that fails either test is a chatbot with a bigger vocabulary.

Will AI replace my employees?

Not the ones doing the part that matters. AI is good at the holding: the tireless, patient attention to detail that keeps every thread alive. It is bad, and should stay away from, the gift: the judgment and warmth your people actually get paid for. In practice it does not empty your team, it moves them off the admin that was quietly eating their week and back onto the work only they can do. If a vendor pitches "replace your staff," they are selling the fantasy that fails in ninety days.

What can AI actually automate in a small business right now?

The reliable wins are the holding tasks: keeping every client record current from real events instead of hand entry, drafting follow-ups grounded in real history, briefing you each morning on what needs you, catching the account that is slipping, recapping calls. Note the shape: it drafts and watches and remembers, and a human approves anything that leaves the building. What it cannot reliably do on its own is the multi-step judgment work, which is exactly why the strongest agents still fail most complex office tasks unaided. Automate the holding. Keep the judgment.


Alex Boquist is the founder of Funal, an AI-run system of work for service businesses. The figures above come from the public sources below, each reviewed directly; everything else is opinion earned the slow way.

Sources