Somewhere in the last two years, the outsourcing conversation changed shape. It used to start with “should we hire offshore?” Now it starts with “should we even hire, or can AI do it?” It is the right question to ask, and most of the answers you will find are wrong in one of two directions: sold by someone with software to sell, or by someone with seats to fill.
We run offshore teams for a living and we put AI tools in front of every person we employ, so we sit in the middle of this question all day. Here is the honest map of it.
What AI genuinely replaced
Start by conceding what is true. A real layer of work that small businesses used to outsource has been absorbed by software, and it is not coming back.
Transaction coding suggestions in accounting software are now good. First draft emails, meeting summaries, document formatting, simple data entry between systems, basic customer question routing: all substantially automated inside the tools businesses already pay for. The purely mechanical seat, where a person’s entire job was moving predictable information from one place to another, is disappearing. If a provider’s pitch is a warm body doing rote work at the lowest possible rate, AI is not their competitor. It is their obituary.
This is why the bottom of the outsourcing market is being squeezed so hard, and it is worth being unsentimental about. Businesses that bought cheap seat filling were buying a commodity, and the commodity found a cheaper form.
What AI conspicuously did not replace
Now the other half of the ledger, which the software pitch skips.
AI does not notice what it was not asked. The bookkeeping model happily codes the anomalous transaction into its most statistically likely category; it takes a person who knows the business to say “that is odd, and I should ask about it.” The gap between plausible and correct is where small businesses actually live, and AI is a machine for producing plausible.
AI does not hold relationships. The patient who calls a clinic anxious about a result does not want a beautifully worded automated message. The client whose invoice query is actually a complaint about feeling ignored needs a human who hears the second thing inside the first. Every small business runs on a few hundred of these moments a year, and they are precisely the moments that decide whether customers stay.
AI does not take accountability. When something goes wrong (and in real operations something always eventually goes wrong) software offers you a log file. A person offers you “that was mine, here is what happened, here is the fix.” Accountability is not a feature that ships in a model update; it is a property of employment, trust and consequence.
And AI does not run itself. The tools need operating: prompting well, checking outputs, knowing which suggestions to accept, keeping the automations honest as the business changes. That is work. Someone does it.
The recomposition: AI capable humans
So the real change is not replacement. It is recomposition. The mechanical layer of admin, bookkeeping and support has compressed; the judgement layer has become the job.
That produces a specific hiring profile: a person who operates AI tools fluently and supplies everything the tools lack: context, scepticism, relationships, accountability. We call this the AI capable human, and the phrase buyers keep reaching for is “human in the loop”, which is accurate as long as you understand the human is the senior partner in that loop.
One person shaped like this now does what two or three purely mechanical roles once did, at a higher standard, because the mechanical layer runs at machine speed while the judgement layer runs at human quality. This, and not cost arbitrage, is the modern argument for a dedicated offshore team member: for the price a local market charges for the mechanical layer alone, you can employ the judgement layer with the mechanical layer included.
But it only works if the person is genuinely trained on the tools, and that is where provider models split. Training people on AI is an investment that pays back over years, so it only makes commercial sense for providers whose people stay years. A provider that churns staff annually cannot rationally train them deeply. Which means the AI question and the retention question turn out to be the same question, a point we have made from the cost side in the real cost of churn in offshore teams.
What this looks like in a real week
Abstractions aside, here is the shape of it in the roles we run every day.
An AI capable bookkeeper lets the software propose the coding, accepts the nine suggestions that are right, catches the one that is wrong because they know the supplier, and spends the reclaimed hours on the work that used to slip: tidy month ends, receivables actually chased, reports the owner reads. The fuller version of that picture, with figures in AUD, is in what offshore bookkeeping costs and how it works.
An AI capable practice receptionist lets the system draft recall messages, then applies what no system holds: this patient prefers a call, that one has been anxious since the diagnosis, this referrer gets cross when letters are late. The clinic feels the difference as fewer gaps in the diary and patients who feel known.
An AI capable web team uses generation tools for scaffolding and first drafts, then applies the craft that decides whether a site is fast, findable and worth reading. The tools make the mediocre version instant. The team makes the good version.
How to decide, for your own business
Strip the vendors out of the decision and it becomes three questions.
First: is the work you want to hand over purely mechanical, with no exceptions, no relationships, no judgement? Then automate it inside the software you already own, and do not hire for it, offshore or anywhere.
Second: does the work involve exceptions, customers, or numbers someone must stand behind? Then you need a human, and the only interesting question is whether that human is AI capable, because a person using the tools well costs the same as a person using them badly and produces roughly twice the value.
Third: will the human still be there when their training has compounded into something rare? A brilliant AI capable person who leaves after eight months takes the compounding with them. The durable advantage is the combination: tools plus judgement plus time.
That combination is not a product you buy once. It is a team you keep.
The next step
If you are weighing software against people for a specific role in your business, put the actual role in front of us. Start the conversation and we will tell you honestly which parts of it AI should absorb and which parts deserve a person, including when the answer is “do not hire anyone, automate it.”