On August 27, 2026, the Bureau of Labor Statistics released a new "AI exposure" classification alongside its 2025-2035 employment projections, sorting occupations into low, moderate, high, and very high categories based on how closely their tasks match current AI capability. Customer service representatives — the occupation closest to answering a business phone — sit among roughly 206 detailed occupations BLS placed in the very-high tier (BLS AI Exposure Categories, 2026; TechInformed, 2026). BLS is explicit that this is not a layoff forecast.
What BLS actually built, and why it's different from the usual "AI will take your job" list
Every prior "which jobs will AI replace" ranking has come from outside researchers guessing at task overlap. BLS's new AI Exposure Categories combine five separate measures into one federal classification: three theoretical scores built by academic economists (Felten, Raj, and Seamans; Eloundou et al.; Eisfeldt et al.), and — for the first time — two measures built from real, observed AI usage: Anthropic's own mapping of actual Claude conversations to work tasks, and Microsoft's mapping of real Copilot usage to work activities (BLS AI Exposure Categories, 2026). That's the shift worth noticing: a federal statistical agency didn't just theorize about which jobs AI could touch, it pulled in telemetry on which jobs AI is actually being used for, right now, by real people.
Where phone-answering work landed — and why, independently, that checks out
Customer service representatives are one of the named occupations in BLS's very-high exposure tier (TechInformed, 2026). That's not a one-off finding. Anthropic's own labor-market research, published directly on its site, independently ranks customer service representatives as the second most-exposed occupation in the U.S. economy by observed usage — behind only computer programmers — and names the specific reason: a large share of real Claude conversations mapped to that occupation are "automated support for payment and billing issues" (Anthropic, 2026). That's precisely the category of call that comes over a small business's phone line every day: where's my order, what do I owe, can I move my appointment.
| BLS category | What it means |
|---|---|
| Low | Task requirements don't align well with current AI capability |
| Moderate | Mid-range alignment between occupational tasks and AI tools |
| High | Substantial portions of the job's tasks can be completed or assisted by AI |
| Very high | Most of the job's tasks match current AI capability; AI performs them frequently in practice |
Exposure isn't a prediction — BLS says so in its own release
BLS's own guidance is direct: "exposure does not imply job loss, productivity gains, automation probability, or wage effects," and a high or very-high rating "does not necessarily mean employment will decline" (BLS AI Exposure Categories, 2026). The same August 27 release backs that up with the actual employment numbers: the U.S. economy is projected to add 5.9 million jobs from 2025 to 2035, growing from 170.3 million to 176.2 million positions. Office and administrative support is the one major group bucking that trend, projected to contract 4.0% — a loss of 752,100 jobs, the steepest decline of any occupational group, which BLS attributes to "the continued integration of automation tools, including those powered by AI, into workflows" (BLS Employment Projections, 2026).
- A real, government-measured decline: office and administrative support, -4.0% (-752,100 jobs) nationally through 2035 (BLS Employment Projections, 2026).
- Not a wipeout: the same decade adds 5.9 million jobs economy-wide (BLS Employment Projections, 2026).
- Not the same measurement: "very high exposure" describes task alignment today, not a forecast of who gets laid off (BLS AI Exposure Categories, 2026).
That distinction is exactly what a wave of "13 jobs AI will replace" roundup posts has been collapsing since the data came out. BLS built a task-alignment scorecard. Most of what's been written about it since reads it as a body count instead.
What it actually means if you're the one who answers the phone
For a solo contractor in Kalispell or a two-location dental practice in Missoula, "customer service representative" isn't an abstract line in a federal spreadsheet — it's whoever picks up when the phone rings, which is often the owner, a spouse, or one employee doing five jobs at once. What this data actually says is narrower and more useful than "AI is coming for the front desk": the specific tasks that dominate real AI phone-answering usage today — appointment questions, order and billing status, routing a caller to the right place — are the tasks a federal agency and an AI company both independently found AI already handles well. The calls that don't fit that pattern — an angry customer, a complicated judgment call — are exactly the ones the data doesn't claim AI has taken over.