Companies Cut Agents for AI. Some Are Already Hiring Them Back.

Last week Gartner published a number I've been waiting for someone to measure.

It surveyed 275 customer service leaders this summer. 21% said their organizations had cut human agents because of AI.

7% said they'd already hired people back.

Another 11% said they plan to within the next 18 months.

Gartner has a name for it: the "cut fast, rehire later" trap.

I don't think it means AI failed. I think a lot of companies cut people based on what AI was supposed to do, not what it had actually done.

Those are two very different things. And the gap between them is expensive.

The headline isn't the whole story

First, the number most coverage will skip.

79% of the leaders Gartner surveyed haven't cut agents because of AI at all. Gartner's own read is that AI's effect on headcount has been modest so far. Hiring freezes, yes. Moving agents into other roles, yes. Mass layoffs, no.

Among the companies that did cut, the average reduction was about 17% of the workforce over 18 months.

So this isn't an industry that fired everyone and is now panicking.

It's a smaller group that moved early. Some of them are finding out the work didn't go away.

Gartner has been predicting that half the companies that cut service staff because of AI will rehire by 2027. This is the first data that looks like it.

Their explanation is the same one I'd give. You cut based on what the AI is expected to handle. It can't handle the volume or the complexity. Service slips, and the people who are left can't close the gap.

Then you're recruiting, onboarding and training all over again. And trying to rebuild knowledge that already walked out the door.

Customers didn't reject AI. They rejected being stuck with it.

In a separate survey of 3,566 customers earlier this year, Gartner found two numbers that look like they contradict each other.

50% said their interactions are easier when companies use GenAI.

87% said it's essential to have an option to reach a human.

They don't contradict. It's the same customer.

Happy to let a bot check an order status. Not happy to be trapped with it when the problem is real.

Picture calling your insurance company about a complicated claim. You've explained it twice. You've uploaded the documents. You've answered the automated questions.

Then the chatbot gives you another generic answer.

At that point you don't care how advanced their AI is. You want someone who understands the problem, can make a decision, and owns it.

Gartner's advice to service leaders was blunt: don't make GenAI "a mandatory first step for every issue." Customers forced through several failed AI attempts before they reach a person are less likely to use the tool again.

That's the backlash. Not the AI. The wall in front of the human.

Klarna is the example everyone points to

In 2024 Klarna announced its AI assistant was doing the work of about 700 full-time agents. It became the poster child for AI customer service.

In May 2025 it started hiring human agents again.

In February, CX Dive covered how CEO Sebastian Siemiatkowski describes where they landed. The AI answers the easy questions. Human service is the premium, "VIP" experience, staffed partly by Klarna's own customers working gig-style. He compared it to driving for Uber.

He also said customers were "frustrated" and "angry" about the 2024 announcement. He called it an eye opener.

Two things I'd take from it.

Klarna didn't give up on AI. It's still automating the simple work. It changed what the humans are for.

And one for anyone in this industry: according to Siemiatkowski, nobody was laid off, because the agents being replaced were outsourced. They moved on to other customer service companies. When a company cuts agents for AI, the outsourced seats are usually the easiest ones to cut. Worth knowing whichever side of the contract you're on.

If you save money on agents and lose customers doing it, that isn't a win.

Cost per contact is the wrong scoreboard

Most AI business cases I see are built on cost per interaction. A bot handles thousands of contacts for a fraction of what an agent costs.

But what happens when customers have to come back two or three times for the same problem?

When they leave a bad review, cancel, or switch to a competitor?

The savings get a lot less obvious.

An automated contact isn't the same as a resolved one. If you want to know whether it's working, look at:

• First-contact resolution
• Repeat contacts on the same issue
• Customer effort and CSAT
• Retention
• Cost per resolved issue, not cost per contact

That last one does the most work. It rarely makes it into the business case. It's the same reason cost alone stopped winning CX outsourcing deals.

Less work for agents isn't the same as no need for agents

Gartner's April survey of 321 service leaders found 85% are adding new tasks and responsibilities to frontline agents as AI takes simple work off their plate. 75% are moving agents into entirely new roles.

That's the part I find most useful.

Once AI handles password resets and order status, what's left for people is the harder stuff. Escalations. Customers about to cancel. A disputed medical bill. A denied claim. A compromised account.

Incomplete information, more than one system, and a customer who's upset. Situations that need judgment, someone who can negotiate, and someone accountable.

The agent's job isn't disappearing. It's getting harder, and more valuable.

Which means you need better agents, not just fewer of them.

Where I'd draw the line

AI earns its place on predictable work. Routing, self-service, scheduling, order status, password resets, knowledge lookups, call summaries, real-time help for agents, quality monitoring.

People still win on complex complaints, sensitive financial and healthcare issues, insurance claims, retention, and technical escalations. Anything where judgment or empathy decides the outcome.

The goal isn't to send everything to a person. It's to know when a person gets a better result, and to make reaching one easy.

And when a customer does move from the bot to a human, the agent should already have the conversation. Nobody wants to explain the same problem three times.

We wrote in July about why 74% of enterprises rolled back an AI agent. Same pattern. The technology usually isn't the problem. The escalation path is.

What this changes if you outsource

Picking a BPO used to mean comparing locations, staffing, pricing and performance.

That all still matters. But there are new questions now, and they aren't side questions anymore.

Who decides which contacts the AI handles? What happens when it can't resolve one? Who's accountable when it gives a customer the wrong answer?

The way I think about it, a good provider should work like a control tower for your customer experience. Watching the AI and the people. Noticing when automation is helping and when it's quietly creating repeat contacts. Adjusting staffing, workflows and quality before your customers notice.

Onshore, nearshore or offshore, same rule. The technology should serve your CX strategy, not set it.

Six questions I'd ask an AI-enabled provider

1. How do you decide what the AI handles and what goes to an agent? You want a method: complexity, risk, customer need, performance data. Not "whatever the bot can contain."
2. What happens when the AI can't resolve an issue? There should be a short, obvious path to a person, and the conversation history should go with the customer.
3. What do you measure besides automation rate? Resolution, repeat contacts, CSAT, retention. If containment is the headline metric, be careful.
4. How do you train agents to work alongside AI? If the easy contacts are gone, every contact left is a hard one. Ask what changed in their training.
5. Who monitors the AI, and who's accountable when it gets something wrong? Ask about QA, compliance controls and escalation rules. Then ask about the last time it happened.
6. Can we change the mix of AI and agents as we go? A good partner gives you room to move. One that pushes a single operating model is selling that model.

You'll learn a lot from which ones they answer easily.

Bringing people back isn't going backward

Some companies worry that adding agents back means admitting the AI strategy didn't work.

I see it differently. Changing course when the results aren't there is just good management.

And it doesn't have to mean rebuilding a big in-house call center. For a lot of companies, outsourced agents are the practical way to restore human capacity without recruiting, training and managing a team from scratch. Dedicated or shared agents, room to flex, people who already know your industry.

Sometimes the answer is more agents. Sometimes it's better automation, a cleaner escalation path, or a more capable partner.

The real question is what's missing from the customer experience you have now.

What I think is going on

AI is going to stay in customer service. It should.

But the companies that come out ahead won't be the ones that automated the most contacts. They'll be the ones that knew which contacts to automate.

Gartner's advice is to treat headcount cuts as the result of AI that has proven itself, not the starting point. I'd put it more simply. Cut after the AI works, not before.

Sometimes the most advanced customer service you can offer is still a well-trained person who picks up, listens, and fixes the problem.

That's a big part of what we do at Outsource Pros.

We work across a network of more than 1,000 vetted contact centers and CX providers, some built around live agents, some around AI, most a mix. We evaluate them from the buyer's side — the things that don't always make it into the sales presentation. Which contacts their AI actually resolves. How fast a customer reaches a person. How they behaved on a program that struggled.

There's no cost to the client.

If you're thinking about bringing agents back, or rethinking how much of your customer service runs on AI, let's talk. We'll look at it with you and tell you what we see.

Sources: Gartner — Survey Finds 21% of Service and Support Leaders Have Reduced Human Agent Headcount Because of AI (Nov 5, 2026) · Gartner — Survey Finds 87% of Customers Say Companies Using GenAI for Customer Service Must Provide Access to a Human Agent (Aug 4, 2026) · Gartner — Survey Finds 85% of Service and Support Leaders Are Expanding Human Agent Responsibilities (Apr 28, 2026) · CX Dive — Klarna pursues 'Uber' style customer service model (Feb 20, 2026)