The AI That Fired People: Klarna and Duolingo, the Two Cases Every Business Owner Must Study | inicIA

Direct answer

In February 2024, Klarna announced that its AI assistant —built with OpenAI— was doing the work equivalent to 700 full-time customer service agents, handling two-thirds of conversations at human-comparable quality and targeting a $40 million profit improvement. The company froze hiring and its headcount fell from about 5,527 to about 3,422 employees over two years through natural attrition. Fifteen months later, in May 2025, CEO Sebastian Siemiatkowski told Bloomberg the cost focus had gone too far —"focused too much on efficiency", resulting in "lower quality"— and Klarna began rehiring humans with an "Uber-style" model of flexible remote agents. Duolingo traced the same arc: in April 2025 CEO Luis von Ahn declared the company "AI-first" and announced it would stop using contractors for work AI could handle; after weeks of public backlash, he clarified he did not see AI replacing his employees, and on the Q2 2025 earnings call admitted user growth landed at the low end of expectations because of the backlash. The lesson for your business: automating to cut costs without protecting quality ends up costing more —in rehiring, in reputation and in customers.

Klarna: from "the work of 700 agents" to "lower quality"

The February 2024 announcement was spectacular: an assistant doing the work equivalent to 700 agents, two-thirds of conversations, human-parity quality, $40M in profit improvement. An honest nuance almost nobody reported: the "700" was a modeled equivalence, not 700 direct layoffs —the headcount reduction came from a hiring freeze and natural attrition, from ~5,527 to ~3,422 employees over two years. The problem was not the number, it was quality: by May 2025, customer complaints and internal reviews showed service had degraded. Siemiatkowski admitted it to Bloomberg and Klarna began rehiring —this time as flexible remote agents, a model one analyst described as completing the circle: they fired to save, spent on AI, and now rebuild the human workforce on worse terms than before.

Duolingo: the "AI-first" that slowed its own growth

In April 2025, Luis von Ahn posted the internal memo on LinkedIn: Duolingo would go "AI-first", "gradually stop using contractors to do work that AI can handle" and only grant new headcount to teams that could not automate more. (In January 2024 it had already cut 10% of its contractors over AI adoption.) The reaction was immediate: users deleting the app after years of streaks, memes and complaints across social media. Weeks later, von Ahn clarified: "I do not see AI as replacing what our employees do (we are, in fact, continuing to hire at the same speed as before)." But the damage was done: on the Q2 2025 earnings call he admitted daily user growth (+40% year over year, vs 60% the prior year) landed at the low end of expectations "because I said some stuff about AI" without enough context, concentrated in the US and Canada among younger audiences. Later, the company stopped evaluating employees on how much they used AI at work.

3 lessons for your business

  1. AI replaces tasks, not judgment. Klarna automated simple conversations well; complex ones needed humans, and that is where quality fell. Automate volume, keep human judgment where the customer notices.
  2. Reversing costs more than doing it right. Rehiring, retraining and repairing reputation costs more than the initial savings. The "savings" of AI-driven layoffs evaporate when quality forces you to rebuild.
  3. The hybrid model wins —and it is communicated honestly. AI for repetitive volume, humans for what requires judgment and empathy. Duolingo's mistake was not the technology, it was the message: its employees were creating 4-5x more content with AI, but the announcement sounded like replacement.

Related articles

Frequently asked questions

Will AI replace my team?

The Klarna and Duolingo evidence says: it replaces repetitive tasks, not entire teams. Companies that laid off for AI and neglected quality ended up rehiring. The model that works is hybrid: AI for volume, humans for judgment.

How do I know which tasks to automate without losing quality?

Start with repetitive, measurable, low-risk tasks (confirmations, scheduling, frequent answers). Keep humans on what requires empathy, negotiation or complex judgment —that is where Klarna lost quality.

What is the hybrid model?

AI + humans working together: the agent handles volume and escalates exceptions to people. Klarna ended up exactly there: AI for the routine and flexible human agents for the rest.

Why did Klarna rehire if the AI "worked"?

Because it worked on efficiency, not quality. Resolution numbers were good; customer experience was not. Its own CEO admitted it: too much cost focus produced "lower quality".

Take the next step

Automate without firing blindly: a well-designed AI agent does the repetitive work and leaves your team for what makes money. Talk to inicIA on WhatsApp and we will design your hybrid model: https://wa.me/17866120930

By inicIA