Most discussions of AI and employment are either catastrophist or dismissive. The catastrophist view holds that AI will eliminate most knowledge work within a decade. The dismissive view holds that AI will create as many jobs as it destroys, as technology always has. Neither is particularly useful for a business leader trying to make decisions today. Three companies that have been unusually transparent about their actual AI-driven changes offer more grounded data.
Klarna's report on its AI assistant deployment is the most cited example. The Swedish payments company reported that its AI customer service assistant, built on OpenAI's technology, was handling the equivalent workload of approximately 700 full-time customer service agents. Klarna's overall headcount decreased significantly as the company reduced its customer service workforce through attrition and did not replace departures. The CEO Sebastian Siemiatkowski was explicit about the connection, stating publicly that the company was doing more work with fewer people as a result of AI. Klarna subsequently walked back some of the more aggressive framing, but the underlying workforce reduction alongside AI deployment was real.
Duolingo's approach is more nuanced. The language learning app reduced its contractor workforce for content creation tasks — specifically the localization and translation work that AI could now perform — while publicly framing this as a shift toward an AI-first content model. The reduction affected primarily contract workers rather than full-time employees. Duolingo has maintained that AI has also enabled content creation at a scale that was not economically possible with human contractors alone, allowing the company to expand into new languages and lesson types faster than before.
IBM took a different approach. CEO Arvind Krishna announced that the company expected to pause hiring for roles that could be performed by AI, primarily in back-office and HR functions, with an estimated 7,800 roles potentially affected over five years. IBM's framing was notable for its specificity: this was not a general hiring freeze but a targeted pause on specific function types where AI substitution was deemed high-probability. Crucially, IBM simultaneously announced expansion in hiring for AI-adjacent roles — the engineers, trainers, and integration specialists building and deploying the AI systems.
The pattern across all three is more specific than the general AI-displaces-jobs narrative: AI is substituting for well-defined, repetitive task categories — customer query handling, content localization, data entry and processing, routine analysis. It is not substituting for judgment-intensive roles, relationship-dependent roles, or roles where novel problem-solving is the primary value. The practical challenge for every organization is accurately categorizing which of their roles fall into which bucket — something that requires more task-level analysis than most job description frameworks are designed to support.