The Overcorrection

A 2025 Orgvue survey found that 55% of companies that laid off employees to replace them with AI now regret the decision. Gartner backed this up in February 2026, predicting that half of companies that cut customer service staff due to AI will rehire by 2027 (often under new titles like "Solution Consultant" or "Trusted Advisor").

The pattern plays out the same way everywhere. Company sees impressive AI demo. CFO runs the numbers on cutting headcount. Leadership decides to "move fast." Agents get laid off. AI goes live. And then reality hits.

What Goes Wrong

The AI handles volume but not variety. Most AI support tools are trained on common questions. The first month looks great: 60% of tickets get handled automatically. But the remaining 40% are the hard ones, and there aren't enough humans left to handle them. Queue times spike. Customer satisfaction drops.

Institutional knowledge walks out the door. That veteran agent who knew that "error 404 on checkout" actually means the customer's ad blocker is interfering? Gone. The agent who could calm down the angriest customer in two sentences? Gone. AI doesn't carry institutional knowledge. It handles patterns it's seen before.

The easy tickets were subsidizing the hard ones. When agents answered a mix of easy and hard tickets, the easy ones gave them breathing room. Remove those (AI handles them now) and every remaining ticket is complex, emotional, or both. Agents burn out faster. The ones who stay get exhausted. The ones who leave are expensive to replace.

Customer expectations don't adjust. Customers don't care that you restructured your support team. They care that they can't reach a human when they need one. And they remember. A Glance study found that 75% of consumers are frustrated by AI customer service. When those consumers have a choice, they choose companies that make human support accessible.

The Companies That Got It Right

Some companies used AI to make their support teams better instead of smaller. The pattern:

Use AI for classification and routing. Messages get categorized and sent to the right team automatically. Agents stop wasting time reading messages that should go to a different department.

Use AI for first-response on simple queries. Password resets, order tracking, business hours: handle these instantly. Agents see only the tickets that need them.

Use AI for agent assistance. Draft suggested responses. Pull up relevant customer history. Summarize long conversation threads. The agent is still in the loop, but they work faster.

Keep headcount stable but shift what agents do. Instead of answering "what are your hours" fifty times a day, agents handle escalations, do proactive outreach, improve documentation, and train the AI on new scenarios.

The Math That Actually Works

Replacing 10 support agents with AI doesn't save you 10 salaries. It saves you maybe 5-6 salaries, because you still need agents for complex issues, AI training, quality assurance, and escalations.

But augmenting 10 agents with AI can double their effective capacity. The same 10 people handle twice the volume because they're not spending time on simple, repetitive queries.

Same or similar cost. Twice the output. No customer satisfaction hit. No rehiring in 12 months.

The Signal for 2026

Gartner's prediction that half of AI-related layoffs will reverse by 2027 isn't speculation. It's based on the data already flowing in from companies like Klarna, and from the Orgvue survey numbers.

If you're evaluating AI support tools, choose ones that work alongside your team, not instead of them. Classification that routes tickets to the right agent. Automation that handles the simple stuff. Assistance that makes agents faster.

The goal is making your team more effective, not making your team smaller.