For most of the AI-layoff cycle, the reassuring story went: the software eats the boring stuff first, then works its way up. Customer service was always the boring stuff. This week the number caught up with the theory. Per Bloomberg, Uber cut about 10% of its customer-service workforce in July as it expanded AI across support operations — the routine tier, the account resets and the “where is my order” tickets, going first.
The frontline was the point all along
Customer support was never a coincidence. It’s the single most automatable white-collar function in a large company: high volume, repetitive scripts, measurable resolution times, and a mountain of past transcripts to train on. If you wanted to build a machine to replace a job, you would design the job to look exactly like a contact center. So the surprising thing isn’t that AI is taking these roles. It’s how quietly the totals are moving.
Microsoft is the cleanest example. Its support workforce reportedly fell from about 50,000 to roughly 40,000 employees and contractors, while sales and service chief Judson Althoff put the AI savings at around $750 million a year as of April 2026. Ten thousand seats and three-quarters of a billion dollars, folded into a slide about “service transformation.” Microsoft’s own Service Agent in Microsoft 365 Copilot went generally available on June 30, with autonomous email resolution slated for September — which is to say the company is selling the exact tool it is using to shrink its own support floor.
Everyone is quoting the same playbook
The pattern repeats with almost comic consistency. The Commonwealth Bank of Australia shed hundreds of chat-support workers, then — after the optics turned — pointed out it had added more than 140 domestic call-center roles over six months, per ABC News. Brinks Home’s CIO said AI cut call volume by roughly two-thirds and the call-center headcount fell from about 800 to 400. In each case the company narrates the same arc: automate the routine tier, keep a thinner bench of humans for escalations and disputes, and describe the whole thing as workers being “freed up” for higher-value work.
The tell is what happens to the people who were freed up. Some transfer, some churn out through attrition, and the seat doesn’t come back. “We didn’t fire them, they moved on” is not a rebuttal to a shrinking payroll — it’s the mechanism.
The offshore hubs are the real earthquake
The number that should make an entire economy sit up isn’t a single company’s headcount. It’s a forecast. Forrester analyst Kate Leggett told Bloomberg that “almost half of customer service roles will be impacted by 2030.” Half. Of a function that employs millions of people worldwide, disproportionately in India and the Philippines, where business-process outsourcing turned answering phones into a national industry.
1-800Accountant expects to spend 50% less on outsourced staff in 2027 after automating simple tasks, and its chief strategy officer Ryan Teeples said the quiet part into a Bloomberg reporter’s recorder: “I wouldn’t want to be an investor in India or the Philippines right now, because that’s where we are seeing the biggest cost savings from AI.” BPO exists precisely because it concentrated the most scriptable work in the cheapest labor markets. That efficiency is now the vulnerability — the more repetitive and well-documented the work, the faster a model absorbs it.
The LostJobs read
There is a survivable version of a support career in 2026, and the companies keep describing it by accident. Every one of them says the same thing: routine goes to the agent, judgment stays with people. The disputed charge, the furious enterprise customer, the case where being wrong costs the company a lawsuit — that still routes to a human, because nobody wants to explain to a regulator that a chatbot approved it.
So the exposure isn’t “customer service” as a category; it’s the tier of it that reads like a script. If your day is resolving the top twenty recurring tickets, the model has already been trained on ten thousand of them. If your day is the messy 5% that the automation escalates — the exceptions, the angry, the genuinely novel — you are on the side of the line that still has a payroll code. The move, as it has been all year, is to get closer to the judgment and further from the script. The scriptable middle is exactly what’s being deleted, and it’s being deleted first.
