AI is the first offshore hire you did not have to requisition

A job changes shape and nobody leaves their desk. That’s the vertigo.

No farewell card. No hiring panel. No new face appearing in a video call from a lower-cost hub.

Just the sentence you’ve typed a hundred times, arriving back cleaner, faster, ready to send.

You didn’t create this bargain by accepting help.

You had a queue, a deadline, a manager waiting for an answer, and a life outside the tab.

There are bills, children to collect, and the small urgent work that accumulates around any job until your afternoon has no spare surface left.

When an assistant offers to take one piece of it, saying yes isn’t a labour strategy.

It’s how a person gets through the day.

Skip the part where the machine develops ambitions.

Notice the procedure instead.

Your employer can now source capacity before admitting that it’s sourcing capacity.

2006 — A role opens in a lower-cost hub. There’s a requisition, a salary band, and somebody arguing about the travel budget.

2015 — The talent search covers three time zones. The candidate has a notice period, a family, and a question about where the work will really happen.

2024 — The assistant fills the first gap in an afternoon. Nobody asks where it lives, because it lives in the browser.

2026 — Nobody raised a requisition. The capability plan changed anyway.

The old hire was visible

A real cross-border hire starts somewhere less magical.

Work.

A security lead isn’t a nationality, a wage or a pin on your map. It’s threat modelling for a named product, release-candidate reviews, incident participation, and mentoring two engineers.

The employer has to name seniority, technical evidence, customer language, salary band, manager, start window, and whether location matters for security, customers or collaboration.

Only then can it claim a shortage.

HR records your domestic search, your compensation benchmark, and the reason for mobility. Your employer compares internal training, a distributed team, a managed specialist for a limited period, and a changed product timetable.

Canada’s Global Skills Strategy aims to process a complete application within two weeks.

A service aim isn’t a product plan.

It still depends on your route being right, your application being complete, and your biometrics arriving on time.

And your offer has to survive more than a manager’s enthusiasm.

The revised EU Blue Card framework accepts a job contract or binding offer of at least six months. Singapore’s Employment Pass route can require forty COMPASS points after the salary stage.

A role can be worth doing and still fail the authorization test.

A permit isn’t a rubber stamp on somebody’s cost saving. It’s one gate in a chain of evidence, conditions and changed facts.

Then comes the part a spreadsheet calls implementation.

Payroll, tax, benefits, identity documents, access, travel, housing, workplace norms, local services, and the actual work.

All of it has to arrive together.

A real plan names your first ninety days. Peers. Product training. Customer-language conventions. Incident procedures. And a check that the job you were promised is the job you were given.

Germany’s EU Blue Card is normally valid for four years unless a shorter contract sets a different calendar. Which is why renewals need an owner rather than a hopeful note in a personnel file.

At thirty, ninety and a hundred and eighty days, your manager reviews outcomes, workload, training, compensation and retention.

Across the OECD, migrants’ average employment rate moved from 71.4% in 2022 to 71.8% in 2023.

That slight change certifies nothing about any hire.

It makes a quieter point.

Admission, employment and durable inclusion are three separate events.

The person was visible at every stage.

So was the dependency.

A country. A manager. A permit. A start date. A payroll record. A renewal calendar. An exit governed by notice and immigration rules.

Your employer could still make a bad decision. It just had to make one out loud, where you could hear it.

The price was never the wage

The old argument for global labour opens with a wage gap, because a wage gap fits on a slide.

Eurostat put 2024 whole-economy hourly labour cost at €10.6 in Bulgaria, €55.2 in Luxembourg, and €33.5 across the EU.

Those are employer-cost statistics. Not prices for one senior bilingual engineer.

They establish a national difference. They don’t settle the cost of the work.

Your fuller price is gross pay, statutory contributions, pension, insurance, bonus, paid leave, recruitment, onboarding, equipment, facilities, management, training, attrition, quality assurance.

A vendor rate has its own concealed contents. Labour, margin, bench, tooling, facilities, risk premium.

The unit that matters to you is one productive hour, or one completed piece of work.

A lower salary that needs twice the coordination, or produces more rework, isn’t the bargain it looked like.

And the gap doesn’t sit still while the contract gets signed.

In 2024, hourly labour costs rose 5.0% across the EU, while Romania and Bulgaria recorded national-currency increases above 13%.

Your hub gets more expensive through local competition, housing, commuting, a shortage of experienced candidates, and the new employer arriving next door.

India can have a very large technical workforce while a 2025 roadmap puts AI-talent supply at half of current demand, against demand rising past 1.25 million people by 2026.

Scale is not the same thing as available skill.

That’s the honest version of arbitrage. An opening rate is a wager on convergence, retention, currency, quality and coordination.

A location portfolio also prices visa restrictions, infrastructure failure, language bottlenecks, mass attrition, duplicate capability, transition time, and what it costs to wind down.

Cheap was never a property of a place. It’s an outcome of a whole operating model.

A hire with no location

Your assistant sidesteps that entire scene.

No domestic search. No binding offer. No relocation provider arranging temporary housing.

No question about whether anybody is isolated, whether the authorization lasts four years, or what happens at a hundred and eighty days.

The capacity appears as a blank box asking you what needs doing.

Which is why “offshore hire” is a useful term and also a wrong one.

The thing in your browser isn’t a migrant worker. And geography hasn’t disappeared.

It moved behind provider infrastructure, model operation, support, data handling, connector design and tool access.

The human being who could once be named in your workforce plan isn’t there.

Your employer’s decision to obtain capacity across a boundary still is.

The agent flow makes that boundary visible if anybody bothers to look.

Looking something up. Drafting a change nobody has made yet. Making the change.

Those aren’t three degrees of cleverness. They’re three degrees of authority.

An agent needs its own machine identity, separate from you and separate from every tool it calls.

Your system should know which boundary failed. Otherwise “the assistant couldn’t do it” becomes a soft sentence covering a hard shift in authority.

None of that is ceremony designed to slow a useful tool down.

It’s the operating model coming back after the interface made it disappear.

Not always a person

Here’s the part that’s hard to say out loud.

The first task your assistant absorbs may not remove anybody.

It may clear your backlog, make a tired person less tired, or let your team meet a deadline that a vacancy had already made impossible.

That benefit is real. Treating every use as a secret redundancy plan would be cruel and lazy.

But capacity release isn’t neutral just because it isn’t a termination.

It changes what gets hired next. Trained next. Retained next. And what you’re allowed to keep doing slowly.

It can make a shortage look solved before anybody decided whether the expertise behind the task should be developed internally, bought from a specialist, kept as a local exception, or allowed to wither.

The labour-market precedent is more honest than the cheerful pitch, because it forces a renewal and exit question.

What happens when the person leaves? What did your employer learn from a bad location assumption, an unrealistic role, a failed relocation, a broken process?

The equivalent agent question was never whether your tab is intelligent.

It’s what happens when the provider changes a model, a connector gains a capability, a memory rule changes, or a tool’s scope quietly expands.

Every material change needs an owner, an evaluation, a rollback plan and a recorded authorization.

An approval isn’t an emoji under a proposed action.

And a committed downstream result, not a confident sentence, is what proves the change happened.

The older theory doesn’t offer you a clean verdict either.

Cost and coordination hazards get inferred, never observed. And firms don’t randomly choose a hub, a supplier, an internal team, or an agent.

Power, distribution, executive preference and office politics don’t become neutral facts by being put in a model.

The record isn’t a calculator. It’s how you refuse the false innocence of the interface.

That’s the correction. You’re not watching geography disappear.

You’re watching the decision to source capacity become quiet enough to miss.

Your employer made help easier to obtain than a conversation about what it would replace.

You didn’t build the arbitrage by accepting the help.

But the old questions are already in the room. What capability is being acquired. Whose work becomes a machine service. What dependency is forming. What authority travels with it.

And who will still understand the work when the helpful answer is the only one that arrives on time.