The Klarna Lesson and 78,557 Laid Off in a Quarter: Why Rebuilding Teams With AI Beats Cutting Them

• Scale of global AI layoffs: 78,557 people in Q1 2026 (48% due to AI) and 88,387 in April (26% due to AI)
• IT market overheating in Russia: vacancies in St. Petersburg down 41%, HH index at 22.9 resumes per opening amid +66% growth in demand for AI skills
• Productivity gap: 2x to 4x between an AI-native specialist and a classically trained employee (+55% speed per GitHub measurements)
• Result of retraining at Alpina Digital: 4.5x acceleration of core production with 0 layoffs

The debate over whether neural networks will replace programmers has, by 2026, moved decisively out of engineering chat rooms and into CFOs' offices. According to Tom's Hardware and Nikkei Asia, the global tech sector laid off 78,557 people in the first quarter of 2026 alone, with 48% of those cuts directly attributed to replacing humans with algorithms. In April, a Challenger report logged another 88,387 layoffs (26% citing AI), while on LinkedIn in the US, the number of new Software Engineer postings dropped 15.6% for the month (Software Developer postings fell 20%).

Russia's labor market is seeing a mirror-image squeeze. In St. Petersburg, the number of IT vacancies shrank 41% year-over-year, and competition on HeadHunter broke through 22.9 resumes per opening in March 2026 — twice the threshold that defines an overheated employer's market. Yet in parallel, Avito Rabota statistics show a 66% year-over-year jump in postings requiring AI tool proficiency.

Why do the polar-opposite cases of

Anthropic and Klarna point to the same management conclusion?

Looking at the two most prominent corporate precedents of the past two years reveals where the technology's limits actually lie.

At one pole stands Anthropic. Back in spring 2025, Dario Amodei predicted that within a year models would be writing almost all code, and in February 2026 Anthropic's CPO Mike Krieger confirmed at the Cisco AI Summit: "Claude now writes itself almost 100%," submitting pull requests of 2,000–3,000 lines. But few notice the other half of the equation: at the same time, Anthropic has massively expanded its headcount of human engineers. When routine syntax-writing moves to the machine, it frees up a huge front of work in architecture, hypothesis testing, and safety.

At the other pole is fintech giant Klarna. In 2024, the company proudly reported that AI had replaced 700 agents and was resolving 75% of inquiries. Yet by early 2026, CEO Sebastian Siemiatkowski publicly admitted the company had overcorrected and resumed hiring humans. The bot handled routine questions well but fell apart on emotionally charged conflicts, nonstandard disputes, and multi-step complaints. The resulting loss of customer loyalty and the cost of cleaning up errors outweighed the payroll savings.

How does role rebundling work, and where are the 4 physical walls of

current models?

Behind the illusion that "tomorrow algorithms will replace everyone" lies a misunderstanding of four fundamental limits of current LLM architecture: the depletion of quality human-written text (synthetic data leads to degradation), the energy ceiling of data centers, the absence of mature quantum computing for modeling complex systems (a 10–15 year horizon), and the disconnection of models from real-world physical sensors.

In practice, then, what wins isn't blind headcount cutting but what analyst Ben Thompson calls rebundling — a deep recomposition of functions within a role. According to McKinsey data (a sample of 600+ companies), deep AI integration into development raises productivity 16–30% and code quality 31–45%, while GitHub measurements show a 55% speedup in closing tasks. Instead of four narrow specialists of the old model, one engineer with an orchestra of agents handles the task.

What should you actually do when hiring and retraining your existing team?

Based on our experience at Alpina Digital and across 40+ client projects, the output gap between an AI-native specialist and an employee still working the old way runs 2 to 4x at the same salary bracket. That leads to three practical steps for leaders:

1. Change your hiring filter: in any interview — from backend developer to lawyer to marketer — replace theoretical questions with a 20-minute practical sprint: build a working prototype or document together with Claude, GPT, or Gemini. 2.

Train your people instead of firing them: in 2024–2025, we ran our entire publishing team through internal intensives. The result was a 4.5x compression of the book-production cycle with zero layoffs: people kept their jobs by rebuilding their own functions.

We teach this exact retraining methodology for engineering and product teams at [Hamidun Academy](https://academy.hamidun.com) and develop it further in weekly workshops at [Vibe Club](https://hamidun.

com/club). 3. Run the math on growth, not just headcount: before cutting 700 positions the way early Klarna did, model the scenario of "keep the team and arm it with AI" to multiply your business's throughput instead.

Developers won't be replaced by a neural network — they'll be replaced by engineers who know how to direct a swarm of models.