Megan Skiendel: Miles, hey — tell me something, did you read that CNBC piece from yesterday?
Miles Ashworth: October twenty-fourth, yes. And I have a position.
Megan Skiendel: Already? We haven't even started.
Miles Ashworth: The position is: one of those two workers is being used. I'm just not certain which one.
Megan Skiendel: Okay, that's actually — look, that's what I want to dig into, because the piece profiles Kaylee Harris and Shannon side by side like they're a tidy before-and-after, same technology, completely opposite experiences. Kaylee's at a credit card company, AI handles her call notes and admin, she says she actually gets to focus on customers now. Shannon's in Cardiff, Wales, telecom rep, and the AI-generated notes are riddled with errors she has to manually fix before she can even start her shift.
Miles Ashworth: Same tool. One person liberated, one person buried.
Megan Skiendel: Exactly — and that's the question this whole thing turns on. It's not whether the technology works. It's why it worked there and didn't there, and honestly I'm not sure anyone's given a straight answer yet.
Miles Ashworth: Well, the straight answer nobody's giving is actually quite simple. Think of autocorrect. When it knows your patterns, your phrases, the way you actually write — it saves you time. When it doesn't, you spend more effort fixing its guesses than you would've spent just typing the thing yourself. That's it. That's the whole mechanism. The AI isn't the variable. The fit between the AI and the specific workflow is the variable.
Megan Skiendel: And Kaylee's company fitted it. Shannon's didn't.
Miles Ashworth: Exactly. Now — Virgin Media O2. They launched an AI voice agent in September 2026. ArvatoConnect is the outsourcing firm running it, and their CEO, Debra Maxwell, set a target: 70% automated interactions, 30% human. That's the ambition on the tin.
Megan Skiendel: And ArvatoConnect claimed 97% accuracy on the AI tool they deployed for a UK government procurement agency. Ninety-seven.
Miles Ashworth: Right — but the part that doesn't fit is Shannon. Cardiff. Telecom rep. Her morning isn't starting calls, it's — actually, no, it's worse than that, it's correcting the previous day's AI notes before she can even think about a customer. Three mischaracterized issues, scrambled account details. That's not a 97% accuracy world.
Megan Skiendel: One of those numbers is wrong. Either the 97% is measured on sanitized test cases that have nothing to do with live calls, or Shannon's deployment is a completely different beast. And look — I've seen both. Vendors demo on their best data. That's not new.
Miles Ashworth: And nobody's rushing to clarify which, frankly. ArvatoConnect hasn't said 'oh, we measured it differently.' CNBC published both claims on October the twenty-fourth, side by side, and there they sit.
Megan Skiendel: Because the contradiction is the story. Kaylee's company invested — actually redesigned the workflow around what the AI could handle. That is the specific decision that made the difference. And it's rare, which means the 70/30 target Debra Maxwell announced is either a genuine plan or a press release, and those are very different things.
Miles Ashworth: And the press release version is the one that gets cited on an earnings call. Which brings us to October the twenty-third — American Express, Q3, the CFO announces roughly five percent of global positions eliminated, three hundred and thirty-six million dollar restructuring charge, and AI expansion named explicitly as the reason.
Megan Skiendel: Named explicitly. On a call with analysts.
Miles Ashworth: And then Marc Benioff does the same thing — four thousand customer-support roles at Salesforce, AI cited. Now, Benioff is — look, he's a specific case, isn't he. Because he profits twice. Once from the Salesforce cuts. And once from selling the software to every other company so they can do the same.
Megan Skiendel: That's not a side effect. That's the product.
Miles Ashworth: Uber cuts ten percent of its customer-support team. Barclays redirects workforce spend straight to AI infrastructure. And I — well, I'm supposed to accept that each of these is independently driven by the technology's sudden maturity?
Megan Skiendel: Look — picture an American Express operations director, fourth quarter looming, the restructuring charge already approved. The AI rollout is underway but honestly it's six months from being reliably good. Does she wait? Or does she take the cut now and let the AI narrative do the work? Because I have been in that room. Not that room, but a room exactly like it. The technology did not drive that decision. The earnings calendar did.
Miles Ashworth: But does the distinction actually matter? If the jobs disappear either way—
Megan Skiendel: Enormously. Because if it's genuine AI efficiency, you can demand companies invest in transition — retraining, workflow redesign, the thing Kaylee's company actually did. If it's just cost-cutting with AI branding, that demand goes nowhere. They'll say 'the technology decided.' And nobody's accountable. That distinction is — actually, it's everything for what we ask of these companies going forward.
Miles Ashworth: And the forecasters who should be keeping them accountable — Gartner says half of companies making these cuts will reverse course. Which means someone already knows the failure rate. They're just not saying it plainly. That number, and what Forrester's predicting for the Philippines BPO sector specifically, is going to make this conversation considerably darker.
Megan Skiendel: Forrester and Gartner are not in the same conversation. Forty-nine percent of customer-service jobs gone by 2030 — that's Forrester. Gartner says no Fortune 500 fully automates by 2028 and half the cutters reverse. Those two firms have presumably looked at the same data and reached conclusions that are — actually, they're not just different, they're structurally incompatible.
Miles Ashworth: Which means the divergence is itself the finding. Nobody knows.
Megan Skiendel: Nobody knows. And the BlueLabel rollout — OpenAI Agents for a fiber provider, halved dispatch calls, eleven thousand dollars a month per role — that's a real number. The help-desk deferral research across nearly forty-eight thousand cases cut costs thirty-eight point one percent. Those gains are real. But they're concentrated in exactly the implementations that had the resources to do it right.
Miles Ashworth: And the Philippines.
Megan Skiendel: USA Today ran a piece on October twenty-third specifically on this — the Philippine BPO sector heading into 2027. Agentic AI is already reshaping staffing and pricing there. That's not a prediction. That's happening to a specific labor market in a specific geography that built its entire economic model on being the place Western companies send their calls.
Miles Ashworth: And Western coverage treats it as a footnote, frankly. Which is — look, that's the Latika M. Bourke warning made concrete, isn't it. She says human customer service becomes a premium tier. Your luxury card gets you Kaylee. Your prepaid account gets the failed loop. Now extend that geographically — the gains flow to the well-resourced implementations, the losses concentrate in one labor market that cannot absorb a single technology wave. That's not an efficiency story.
Megan Skiendel: No, it's a distribution story. And if Gartner is right — if half these companies reverse course — they'll rehire, but not in Manila. They'll rebuild domestically, cheaply, probably on shorter contracts. The Philippine workers don't get a reversal.
Miles Ashworth: So the Gartner scenario is actually worse for them than the Forrester one.
Megan Skiendel: In one version the jobs disappear permanently and that's the story. In the other version, Western companies stumble, reverse, quietly rebuild — and a concentrated labor shock in one geography just gets quietly written off as a transition cost. The forecasts disagree on the scale. They agree on who absorbs it.
Miles Ashworth: Which is a quietly devastating place to end up, frankly. We started with two workers. Kaylee and Shannon. Same technology, the piece said. And that framing — same technology — was the thing that was always slightly wrong, wasn't it.
Megan Skiendel: It was never the same technology. It was the same category of technology deployed by people who made completely different choices about what mattered. Kaylee's company chose implementation. Shannon's company chose the announcement. And the next earnings call from a major financial services company — American Express, whoever's next — that call will tell us which of those two futures is actually winning. Not what they say. What they do in the twelve months after the cuts. If Gartner's right and half of them reverse, the rehiring is the tell. Watch who they hire, where, on what contracts.
Miles Ashworth: One of those two workers is being used. That was my position at the top. I think they both are, actually. Just — differently.
