AI Is Replacing Customer Service — Except the Part That Actually Matters

Anthropic just published research showing customer service reps are the second-most AI-exposed profession in America. Seventy percent of the job can theoretically be automated. This is especially relevant when thinking about AI replacing customer service.

Also, and honestly? They’re right. Most customer service is text. Tickets, chat windows, scripted responses, knowledge base lookups. An AI agent can handle “how do I reset my password” better than a human, faster, more consistent, available at 3 AM on a Saturday.

So here’s the uncomfortable question nobody in the industry wants to ask: If 70% of customer service can be automated, what’s in the other 30%?

Furthermore, that’s the interesting part. And it’s where the entire “AI replaces support” narrative falls apart.

The 70% That’s Already Dead: Understanding AI replacing customer service

Moreover, let’s be honest about what AI is good at in customer service:

• Answering FAQs. If the answer exists in a knowledge base, an AI chatbot will find it faster than any human. This was true five years ago with basic search. It’s embarrassingly true now.
• Routing tickets. Categorizing, prioritizing, and assigning support requests. AI does this better than rules-based systems, and infinitely better than a human reading every ticket.
• Generating responses. For standard issues with standard resolutions, AI can draft (or auto-send) perfectly adequate responses. The customer doesn’t know. The customer doesn’t care.
• Summarizing conversations. After a support call, AI can produce a case summary in seconds. This used to take agents 5-10 minutes per interaction.

Furthermore, if your support team’s primary value is typing answers to common questions, AI isn’t “coming for your job.” It already has it. The companies that haven’t automated this yet are just slow.

The 30% AI Can’t Touch

Additionally, here’s what Anthropic’s exposure metric doesn’t capture: the difficulty isn’t uniform across that remaining 30%. The tasks AI can’t do aren’t just “harder”, they’re in a fundamentally different category.

They require seeing the real world.

In addition, when a customer calls because their industrial printer is making a grinding noise, no chatbot on earth can diagnose the problem from a text description. When a homeowner’s HVAC system is leaking and they can’t describe where, because they don’t know HVAC terminology, the solution isn’t a better language model. It’s eyes on the problem.

They require real-time judgment in ambiguous situations.

The customer says “it’s not working.” The AI asks clarifying questions. The customer says “I don’t know, it just looks wrong.” Now what? A human support agent can say “can you show me?” and make a judgment call in seconds. An AI can only operate on information that’s been converted to text. And most real-world problems resist that conversion.

They require trust at the moment of crisis.

When something expensive is broken, a medical device, a piece of equipment, a home system, the customer isn’t just looking for information. They’re looking for reassurance from another human being that the problem is going to get solved. AI can simulate empathy. It cannot provide the genuine human connection that de-escalates a panicking customer. Not yet. Maybe not ever.

The Dangerous Middle Ground

Here’s what I think is actually happening, and it’s more nuanced than “AI good” or “AI bad”:

Companies are automating the easy 70% and then expecting the remaining 30% to just… figure itself out.

They lay off half the support team because chatbots handle ticket volume. Then the complex cases, the ones that actually require human judgment, visual diagnosis, or real-time problem-solving, pile up. The remaining agents are overwhelmed. Resolution times for hard problems get worse. Customer satisfaction on easy issues goes up (faster AI responses). Customer satisfaction on hard issues craters.

The aggregate CSAT score might stay flat. Executives call it a win. But the customers with the hardest problems, who are often the highest-value customers with the most complex products, are getting worse service than before.

This is the blind spot. And it’s not a technology problem. It’s a strategy problem.

What Smart Support Teams Are Actually Doing

The support teams that are getting this right aren’t choosing between AI and humans. They’re doing something more interesting:

They’re using AI to handle the 70%, and they’re investing the savings into making the 30% dramatically better.

That means equipping agents with tools for the hard problems:
– Visual support, letting agents see what the customer sees through their phone camera, without requiring an app download
– Better diagnostic tools, giving agents real-time data about the customer’s product or account before the conversation starts
– Specialist routing, using AI to identify complex cases early and route them to experienced agents instead of tier-1 generalists

The teams that treat AI as a way to elevate human support (rather than replace it) are the ones building competitive moats. Because here’s the thing: every company can deploy the same chatbot. The 70% becomes commoditized. The 30% is where you differentiate.

The Question Nobody’s Asking

Anthropic’s research found “limited evidence that AI has affected employment to date.” The gap between what AI can do and what it’s actually doing in the real world is still massive. Legal constraints, technical limitations, and the simple friction of organizational change all slow adoption.

But here’s the question I keep coming back to:

When every company automates the same 70% of customer service, what becomes the competitive advantage?

It’s not the chatbot. Everyone will have one. It’s not the ticket routing. That’s a commodity feature.

It’s what happens when the chatbot can’t solve the problem. Additionally, it’s the moment a customer needs a real human to look at their broken thing and help them fix it. It’s the 30% that AI can’t do, the part that’s messy, visual, ambiguous, and deeply human.

That’s where support teams earn their keep. And that’s where the investment should be going.

If your team is thinking about how to handle the 30% that AI can’t automate, we built something for that.

For additional context, see recent analysis from Gartner research on trends in this space.