What we do that AI can’t

Every few days, someone asks a version of the same question. Usually, it comes from an entrepreneur who has discovered a new AI tool, seen what it can do in a matter of seconds, and started wondering whether they still need a VA or an Online Business Manager at all. My LinkedIn feed and even some of the networking groups I’m in seem full of people who think that AI can do everything their team used to.

It is a fair question, and I think it deserves a proper answer rather than a defensive one.

AI is genuinely good at parts of this job. It can draft a perfectly serviceable email, summarise a meeting, format a document, pull together a first-pass itinerary or turn a rough brief into something structured and usable. It can speed up research, reduce repetitive admin, and remove some of the lower-value work that has traditionally taken up far too much time.

And that is a good thing.

If your idea of business support is simply having someone carry out a list of tasks on request, then yes, AI is changing that part of the market quickly. But I would argue that task execution was never the most valuable part of good business support in the first place.

The task was never the real value

For years, support work has often been described in terms of the tasks being completed. Book the flight. Send the invoice. Draft the post. Update the spreadsheet. Organise the agenda.

Those are the visible outputs, so naturally they are the things people notice. But the real value is usually found in the thinking that happens around them.

Booking a flight is one thing. Knowing which flight makes sense because there is a client call in another time zone later that day is another. Sending an invoice is straightforward. Knowing that it needs to go before a particular deadline, or that a client has changed their usual payment pattern and may need following up, requires context. Dealing with a “quick favour” is easy enough in isolation. Recognising that it is the third quick favour from the same client this month, and that it may be time to address scope or boundaries, is something else entirely.

That is the part that is easy to overlook because it rarely appears on a task list. It is judgement. It is pattern recognition. It is knowing the person, the business and the wider context well enough to understand what matters before simply carrying out what has been asked.

Recently a client asked me to help manage their inbox, but they were already using an AI tool to automatically label invoices.  I noticed that every notification of a sale was being wrongly labelled as an invoice which meant that actual invoices were being drowned.  This example shows how useful AI can be, but also why it still needs human oversight to make it genuinely effective.

AI is becoming very good at producing the visible output. What it does not automatically have is the history behind the request, the nuances around it or the understanding of why one option may be better than another for this person, in this particular business, at this particular moment.

Good support goes beyond the request

A tool responds to what you ask it. Good support often begins with what you did not think to ask.

That might mean noticing a clash in the diary before it becomes a problem, double-checking a booking because something about it does not look quite right, or remembering that a client prefers things handled in a particular way. It might mean seeing that someone has overloaded their week again, despite saying they wanted more breathing space, and quietly helping to create some before everything starts to feel unmanageable.

Over time, when you work closely with someone, you learn far more than their systems and processes. You learn their habits, their pressure points, the things they tend to overlook and the situations that create unnecessary stress. You begin to understand what they will probably regret not dealing with now, even if it does not yet feel urgent to them.

That context accumulates slowly. It comes from being inside the business, seeing patterns repeat and understanding not only what someone says they need, but how they work in reality.

Diary management is an example of this.  AI can absolutely help schedule appointments, providing it’s been carefully prompted beforehand. Do you need a lunch break every day, to finish at a certain time in the evening, and a breather between meetings?  We are not machines, and AI won’t know that you had a bad night’s sleep and need a gentler day than the one you’d initially planned.

This is where anticipation becomes so important. Good support is not simply reactive. It is offering more than one solution when there are several ways forward. It is asking the extra question. It is catching something before it becomes visible as a problem. It is knowing when to act, when to flag something and when to push back.

That kind of judgement is built through trust and familiarity. It is not generated in a single interaction.

Working across languages also means working across systems

For multilingual entrepreneurs, there is another layer to this.

Translation software is extremely useful, and AI can move words from one language to another with impressive speed. But language is only part of the picture. The systems, expectations and cultural context surrounding those words matter just as much.

A French administrative deadline and a British commercial expectation may operate according to completely different logic. A sentence may be technically correct in both languages while carrying a very different implication in practice. A piece of correspondence may appear straightforward until you understand what usually happens next, what has been left unsaid, or what the other party is expecting from you.

Anyone who has dealt with French administration knows there is often a significant difference between understanding the words on the page and understanding what you need to do.

Whilst the UK might have clear government process outlined on a plethora of government websites, the reality on the ground here in France can vary enormously from one prefecture, and sometimes one person, to another.  
That is not really a translation problem. It is a context problem.

There are things you understand because you have dealt with them before. You have made the call, completed the form, sat in the queue, spoken to the person at the other end and learned what the process looks like in reality rather than in theory. Knowing that ‘no’ isn’t always the final answer is one of the most important lessons any expatriate can learn.

That kind of knowledge matters because it shapes the decisions you make. It helps you know what needs checking, what needs questioning and when something that looks perfectly normal on paper is likely to become a problem further down the line.

Accountability is not a feature

There is another difference that matters just as much, and perhaps more than any of the others: accountability.
When something goes wrong, an AI tool has no stake in the outcome. It does not worry that a deadline has been missed. It does not notice that a payment has not arrived and decide to follow it up. It does not think about a client conversation later and wonder whether there was something underneath it that needs addressing.

A person does.

When you work closely inside someone’s business, you care about what happens next. You remember what is outstanding. You notice when something feels off. You follow up. You take ownership of your part in the work, and when something needs fixing, you help fix it.

That sense of responsibility is difficult to reduce to a deliverable because it sits underneath everything else. It is part of the relationship itself.

When we plan travel for our clients, we try very hard to put ourselves in their shoes: what do they need from the moment they leave their front door, to the moment they return to it? For many business owners, that is what they are really buying when they bring the right support into their business. It is not simply another pair of hands. It is knowing that someone else is paying attention.

This is not about humans versus AI

None of this means AI has no place in good business support. Quite the opposite.

Used well, it can make the work faster, cleaner and more efficient. It can remove repetitive tasks, help create a useful first draft, organise information and reduce the amount of time spent on things that do not need deep human involvement.

That creates more space for the work that does.

At The Bilingual Lantern, we use AI on a daily basis, and it enables us to be more efficient and productive than we could ever be without it. However, I won’t hand over the reins completely. I know business owners around me who blindly hand over their inbox, agendas and files to a virtual agent and leave it to manage everything, but I honestly don’t want to go down that road. Quite apart from the confidentiality and data-protection questions that raises, my clients don’t just pay me to get tasks done, they pay me for the relationship I have with them, their clients and their suppliers.

I do not think AI and excellent business support are competing for exactly the same job. In many ways, they were never doing the same job to begin with.

The important question is not simply, “Can AI do this task?” In many cases, the answer is yes.

The better question is, “Who is thinking about my business when I am not in the room?”

Who is holding the wider context? Who notices the pattern? Who spots the risk? Who remembers the thing you mentioned six weeks ago? Who understands that a perfectly sensible solution on paper may not work for you in practice?

That is where human support still matters.

And, for me, that has always been the point.

If you’re a busy entrepreneur, AI can be a gamechanger, but it needs context. It needs to understand who you are, your mission and your values are, the purpose your business serves and who your clients are, before it can be genuinely useful.

If that is the kind of support you are looking for in your business, someone who is not simply waiting for the next task, but is paying attention to the whole picture, then we would be very happy to talk.