AI Layoffs and the Productivity Story in 2026: The Klarna Walk-Back and What It Tells Us
Klarna walked back its AI agent claims. Salesforce paused hiring on Agentforce. IBM, Cognizant, Infosys, and the consulting firms all cite AI exposure. The story is more complicated than headline.
The clearest data point in the 2024-2026 conversation about AI and white-collar employment is the Klarna walk-back. In February 2024 Sebastian Siemiatkowski announced that the company’s customer-service AI agent — built on OpenAI — was doing the work of 700 full-time agents. The claim became the most-cited example of generative AI displacing labor and was referenced in dozens of analyst reports and board decks through 2024. By mid-2025 Klarna was quietly rehiring for customer-service roles and the CEO was on record acknowledging the agent had degraded the customer experience in ways the metrics had not captured.
The Klarna story is the cleanest illustration of a broader pattern. The headline AI displacement claims are mostly louder than the underlying reality. The productivity gains are real but uneven, the substitution is partial, and the labor-market attribution is harder than press-release language suggests. This is what the actual picture looks like in 2026.
The Klarna walk-back, in detail#
The original February 2024 claim was that the OpenAI-powered Klarna assistant handled two-thirds of customer-service conversations in its first month and was doing the work of 700 agents. The reported metrics included higher CSAT than the human baseline, faster resolution times, and a projected 40 million dollar profit improvement in 2024.
The walk-back came in stages through 2024 and 2025. The first signal was a Bloomberg piece in late 2024 reporting that complex queries were degrading in resolution quality and that Klarna had been re-staffing in specialized roles. The second was the Q1 2025 earnings call where Siemiatkowski himself acknowledged that the customer experience had suffered and that the company was investing in human-and-AI hybrid roles. The IPO marketing in late 2025 emphasized the hybrid model rather than the displacement story.
The lesson is not that AI customer-service agents do not work — they do, for the bounded portion of the query distribution where they work. The lesson is that headline displacement numbers tend to over-attribute simple-query automation to total-workload substitution.
Salesforce, Agentforce, and the hiring freeze#
Salesforce’s Agentforce launch in September 2024 was the largest enterprise pivot to AI agents from a major SaaS vendor. The company announced an explicit hiring freeze on customer-facing engineering roles in early 2025, with CEO Marc Benioff publicly attributing the decision to AI-driven productivity. The headline was that Salesforce had stopped hiring software engineers because Agentforce and the internal AI tooling could absorb the workload.
The reality was more nuanced. The hiring freeze was selective — Salesforce continued hiring AI engineers, ML researchers, and certain product roles — and the productivity argument was contested internally. Mid-2025 reporting from The Information and Bloomberg surfaced that the freeze was as much a margin-driven move as an AI-driven one, with Salesforce facing pressure on growth-versus-profitability tradeoffs from activist investors.
By 2026 Salesforce had selectively reopened engineering hiring in specific groups while maintaining the Agentforce narrative externally. The CRM market dynamic — Salesforce defending its position against ServiceNow’s Now Assist, Microsoft’s Copilot push, and the broader vertical-SaaS challenge — is the more material driver than the AI productivity story.
IBM and the 7,800 headcount figure#
IBM’s announcement in 2023 that 7,800 back-office roles would be candidates for AI-driven displacement over five years became the most-quoted figure in the enterprise AI-and-labor conversation. The specific roles named were in HR, finance, and other back-office functions; the timeline was deliberately extended.
The IBM 2024 and 2025 actual headcount changes told a more complicated story. Total IBM headcount declined modestly through 2024 and into 2025, with the largest reductions in consulting and certain legacy services lines. Back-office automation through watsonx Orchestrate has delivered measurable productivity gains in specific HR and finance workflows, but the 7,800 figure has not appeared as a clear net reduction in the equivalent functions. The pattern looks more like: AI absorbs growth that would otherwise have required hiring, rather than displacing existing roles outright.
The IBM-Cognizant data point matters for a related reason. IBM’s services business and the major IT-services firms (Cognizant, Infosys, Wipro, TCS, Accenture) are the part of the labor market most directly exposed to AI productivity in the near term, because their billable work is heavily code generation, ticket handling, and document production.
Cognizant and Infosys: the IT services compression#
Cognizant announced workforce restructuring in 2024 and 2025 with explicit reference to AI productivity. The company’s net headcount declined modestly through 2025 while revenue continued to grow — the classic “doing more with less” pattern that AI proponents cite.
Infosys’s posture has been similar. The company’s Q4 FY2025 commentary referenced AI productivity contributing to lower fresher hiring and a more selective lateral pipeline. TCS, the largest of the Indian IT services firms, has maintained higher headcount growth but with explicit deployment of AI tooling across its delivery teams.
The IT services pattern reflects a sector-specific dynamic. The work is heavily code-generation and ticket-resolution, which is genuinely where AI productivity is strongest. The clients (US and European enterprises) are willing to pay similar prices for AI-augmented delivery while expecting the volume of delivered work to increase. The net effect is selective hiring slowdown and aggressive productivity-tool deployment — not mass layoffs.
The fresher-hiring slowdown matters disproportionately because the Indian IT services industry has historically been a major engineering-graduate employer. The 2024 and 2025 reductions in fresher hiring (the entry-level graduate intake) have been the most-discussed labor-market consequence in India.
The consulting story: McKinsey, BCG, Bain#
The strategy consulting firms have been the most surprising entrants in the AI-labor conversation. McKinsey announced workforce reductions through 2024 — roughly 3 percent of the global workforce — explicitly citing efficiency gains from AI tooling. BCG and Bain followed with more modest reductions and similar framing. The headline was that even strategy consulting, the white-collar work of white-collar work, was AI-exposed.
The reality is the consulting industry was running into a 2023-2024 demand softening that pre-dated AI productivity. Project pipelines softened. The McKinsey Solutions group had been over-staffed during the 2021 boom. The AI productivity story was the public-facing rationalization for what was substantially a normal cyclical contraction. The actual deployment of AI tooling inside the firms — slide generation, research summarization, document drafting — is real and material, but the headcount reductions over-attribute to it.
What actually attributes to AI#
The honest analytical picture across 2024-2026 is that AI productivity is contributing to slower hiring rather than active displacement, in specific functions, with significant variance by industry. The functions where the pattern is strongest are software engineering (Copilot, Cursor, internal AI tooling absorbing roughly 20 to 35 percent of routine code generation), customer service (the Klarna pattern, with the partial-substitution caveat), entry-level marketing copy and content production, and certain back-office functions (accounting close, expense processing, basic legal document review).
The functions where the pattern is weaker than headlines suggest include senior engineering roles, complex customer service, strategic decision-making, and creative work above the routine layer. The labor-market signal is selective hiring slowdown at entry levels, deferred backfilling of attrition, and aggressive tooling investment — not the wave of displacement the 2023 forecasts predicted.
The macro view#
The aggregate US white-collar labor market through 2024 and 2025 has been softer than the historical pattern but not collapsing. The unemployment rate among college graduates has risen modestly. Entry-level technology hiring has slowed materially. The labor-market signal is real but the attribution is mixed — interest rates, post-2021 over-hiring corrections, and AI productivity all contribute.
For 2026 the consensus among labor economists who work on this question (Goldin and Katz, Acemoglu, the MIT Future of Work group) is that the displacement projections in 2023 were premature. The actual rate of AI-driven labor substitution is slower and more selective than the 2023 estimates suggested. The longer-run picture remains genuinely uncertain.
Where pdpspectra fits#
We build AI systems that improve productivity inside enterprises without the over-claiming. Our business automation practice deploys agent workflows, document processing, and code-generation tooling with realistic measurement of productivity impact and honest assessment of where human work is still required.
Related reading: tech layoffs and what they mean 2025-2026, enterprise AI rollout roadmap, and AI agents back-office automation.
The AI-and-labor story is more complicated than headline. Talk to our team about an AI deployment that measures productivity honestly and avoids the Klarna trap.