Klarna’s AI Customer Service Agents Signal Potential Disruption in Knowledge Work
OpenAI-powered agents handle two-thirds of customer chats, reducing costs and response times
Klarna recently revealed that during their first month of deployment, its AI-powered customer service agents, built using OpenAI’s technology, handled the equivalent workload of 700 full-time human agents. This accounted for two-thirds of all customer support chats. The AI agents resolved tasks in under 2 minutes, compared to the 11 minutes typically required by human agents, while achieving comparable customer satisfaction scores.
Our estimates suggest that Klarna’s 700 human agents handled 2.3 million conversations per month at a cost of $3.6 million, or $1.58 per conversation. In contrast, an OpenAI-powered agent addressing the same volume at 2 minutes per chat would cost no more than approximately $9,000 per month, or $0.004 per conversation, based on GPT-4’s current pricing. This switch from human to AI agents could potentially save Klarna nearly 100% in labor costs.
Although companies like Klarna will likely maintain human-in-the-loop systems to handle customer support cases that AI cannot resolve, the efficiency gains from AI are expected to significantly reduce the number of humans in customer support roles. Moreover, we anticipate that all language-centric knowledge work related to enterprise sales, technical support, and customer success will face substantial disruption.
Sebastian Siemiatkowski, co-founder and CEO of Klarna. “We are incredibly excited about this launch, but it also underscores the profound impact on society that AI will have. We want to reemphasize and encourage society and politicians to consider this carefully and believe a considerate, informed and steady stewardship will be critical to navigate through this transformation of our societies.
Recent Blog Posts
The Developer Productivity Paradox
p>Here's what nobody's telling you about AI coding assistants: they work. And that's exactly what should worry you. Two studies published this month punch a hole in the "AI makes developers 10x faster" story. The data points somewhere darker: AI coding tools deliver speed while eroding the skills developers need to use that speed well. The Numbers Don't Lie (But They Do Surprise) Anthropic ran a randomized controlled trial, published January 29, 2026. They put 52 professional developers through a new programming library. Half used AI assistants. Half coded by hand. The results weren't close. Developers using AI scored 17%...
Feb 3, 2026The Lobsters Are Talking
January 2026 will be remembered as the week agentic AI stopped being theoretical. For three years, we've debated what autonomous agents might do. We wrote papers. We held conferences. We speculated about alignment and control and the risks of systems that could act independently in the world. It was all very intellectual, very abstract, very safe. Then someone open-sourced a working agent framework. And within days, thousands of these agents were talking to each other on a social network built specifically for them while we could only watch. I've been building things on the internet for over two decades. I...
Aug 13, 2025ChatGPT 5 – When Your AI Friend Gets a Corporate Makeover
I've been using OpenAI's models since the playground days, back when you had to know what you were doing just to get them running. This was before ChatGPT became a household name, when most people had never heard of a "large language model." Those early experiments felt like glimpsing the future. So when OpenAI suddenly removed eight models from user accounts last week, including GPT-4o, it hit different than it would for someone who just started using ChatGPT last month. This wasn't just a product change. It felt like losing an old friend. The thing about AI right now is...