On the night shift, the Philippines' call centres meet their replacement

MANILA — It is just past eleven at night in a tower in Bonifacio Global City, and the floor is awake the way only a Philippine call centre is awake at this hour: bright, caffeinated, and synced to the business day of a country eight thousand miles west. Rosario, a 29-year-old agent who has worked the American overnight shift for six years, settles her headset and signs in. Beside her name on the screen, something new is waiting.

It is an AI co-pilot — a system that listens to her calls in real time, suggests answers, drafts her after-call notes, and silently scores how closely she follows the script. Six months ago it did not exist. Now it is, in effect, her colleague, her assistant and her supervisor, all at once, and she is still deciding how she feels about it.

The Philippine outsourcing industry employs well over a million people and has been, for two decades, the surest ladder out of precarity for English-speaking young Filipinos. It is also, by common consent, among the industries most exposed to artificial intelligence anywhere on earth. On this floor, at this hour, the abstract debate about AI and jobs is not abstract. It is wearing a headset.

The co-pilot on the line

The technology arrived gently, which is how these things usually arrive. Management framed the co-pilot as help, not threat: a tool to cut the dreaded after-call work, to surface answers faster, to make hard calls easier. For Rosario, parts of it genuinely are help. The system drafts her notes in seconds, sparing her the few minutes of typing that used to follow every call and stack up across a shift.

But the same system that helps her also watches her. It measures her adherence to the script, flags when she goes off-process, and feeds a stream of data to supervisors who once relied on spot-checks and intuition. The assistant and the monitor are the same software, and the agents are under no illusion about which role matters more to the people who bought it.

It listens to everything. Some nights it feels like having a manager inside your headset who never sleeps and never likes you, said one agent, a five-year veteran, asking that her name not be used.

The productivity gains are real, and that is precisely the worry. If each agent, augmented by AI, can handle more calls and more complexity, then the same volume of work needs fewer agents. The co-pilot makes the survivors more valuable and the marginal hire less necessary, and everyone on the floor can do that arithmetic.

Where the cuts land first

The exposure is not evenly spread, and the floor knows the hierarchy of risk by heart. The most vulnerable work is the simplest: password resets, balance enquiries, order tracking, the scripted exchanges that AI handles end to end with no human at all. That tier is already thinning, as automated systems intercept the easy calls before they ever reach a person.

What is left for the humans is the hard stuff — the angry customer, the tangled complaint, the situation that requires judgement, empathy or improvisation the machine cannot manage. That is, in a sense, more interesting and more skilled work. It is also fewer jobs, paid to a smaller, more capable group, with a higher bar to clear to be in the room at all.

Industry executives insist the headcount is holding for now, and point to a shift in the mix toward higher-value services rather than a collapse in numbers. The agents on the floor are more sceptical. They have watched the easy queues shrink, the targets climb, and the new-hire classes get smaller. Nobody has announced a wave of layoffs. Everyone senses the ground moving.

The ladder that built a middle class

To understand the stakes, you have to understand what these jobs mean here. For two decades, a BPO headset has been the most reliable way for a young Filipino with good English and no connections to earn a stable, formal wage — to support parents, send siblings to school, escape the informal economy. The industry did not just create jobs. It created a pathway into the middle class that little else offered.

That pathway is now narrowing at exactly the rung where it was widest. The entry-level, low-complexity roles that absorbed waves of new graduates are the ones AI eats first. If those disappear without something to replace them, the country loses not just employment but a social escalator that an entire generation has depended on.

Rosario is one of the lucky ones, and she knows it. Her tenure, her skill with difficult callers and her willingness to work nights make her the kind of agent the new model wants to keep. The friends who started alongside her on simpler accounts are less sure of their footing, and the conversations on the smoking deck have changed from gossip to quiet calculation.

Climbing or clinging

The official answer to all this is reskilling, and it is not entirely empty. Firms and government agencies talk of moving up the value chain — into analytics, software support, healthcare administration and other work that is harder to automate and pays more. Some agents will make that climb, and the industry's leaders are betting the sector reinvents itself faster than AI hollows it out.

But reskilling is easier in a slide deck than on a night shift. Not every agent can become a data analyst, and the new higher-value roles will not, by their nature, employ as many people as the call queues did. The optimistic case is not that everyone is saved. It is that the industry survives, smaller and smarter, while many of the people who built it are quietly left behind.

At a quarter past midnight, Rosario takes another call — a furious customer, a billing dispute, the kind of human mess the co-pilot can assist with but not resolve. She talks the man down, fixes the problem, and lets the AI draft her notes while she breathes. For tonight, the division of labour holds: the machine handles the easy and the routine, and the human handles the hard. The open question, on this floor and across this industry, is how much hard work will be left to go around — and how many headsets will still be lit when the answer arrives.