Will AI Replace Developers — and What Companies Should Do About It

In Q1 2026, tech companies laid off 80,000 people — almost half of those layoffs were caused by AI. Klarna has already wound down its experiment of replacing human staff with AI and is calling employees back. What should a B2B company do before repeating others' mistakes.

Zhemal Khamidun, Head of AI Alpina Digital, CPO AlpinaGPT

Below are the key figures I would recommend reviewing before moving forward. Per data from Tom's Hardware and Nikkei Asia, tech companies laid off 78,557 people in Q1 2026 — and nearly 48% of those layoffs were directly tied to replacing human workers with AI and automation. This share rose in April: per the Challenger report, AI was the root cause of 26% of the 88,387 monthly layoffs that month. The situation is nearly identical in Russia: the number of IT vacancies has dropped by 41% year-over-year, and the HH index for the IT sector has reached 22.9 resumes per open vacancy.

Against this backdrop, the debate over whether AI will replace developers has shifted from informal IT team chats to the CFO's office. The question has changed accordingly: what should a company do before replicating Klarna's experiment, where after mass AI-driven layoffs in 2024, the CEO publicly admitted the move was a mistake and reopened hiring in early 2026.

My name is Jemal Hamidun, I am the CPO of AlpinaGPT and Head of AI at Alpina Digital, and the author of the Telegram channel «Cooking Up AI». Over two years, we have carried out AI transformation for our own publishing house and 40+ corporate clients — pharma, retail, fintech, and media. In this article, I have compiled everything we learned about hiring, layoffs, and team retraining in the new environment, based on specific case studies and figures.

Anthropic on one side of the barricade

In March 2025, Anthropic CEO Dario Amodei at the Council on Foreign Relations said outright: "In three to six months, we will be in a world where AI writes 90% of code, and in twelve — practically all code." That sounded bold. In February 2026, Anthropic CPO Mike Krueger at the Cisco AI Summit confirmed: "Claude is now writing Claude — this is effectively 100%." Internal Anthropic reports show pull requests of 2,000–3,000 lines, fully generated by the model.

There is one key detail here that gets lost in retellings. At the same time, Anthropic continues to aggressively hire engineers — the team size has grown severalfold over the same period. This creates a counterintuitive setup for casual observers: AI writes the code, but more people are needed. This paradox is the main takeaway any company needs to understand before drawing arrows to indicate layoffs.

Klarna on the other side

The second pole is Klarna. In 2024, CEO Sebastian Siemiatkowski announced that AI had replaced 700 support agents and handled 75% of all customer chats. The story spread across the world as a model of "how it should be done". It was only in 2026 that this story gained a second, far less discussed chapter.

By the start of 2026, Klarna quietly rolled back the experiment and began rehiring staff. This was not because the AI proved ineffective — it handled standard queries perfectly and kept up with demand. It failed to cope with non-standard cases: emotional conflicts, complex disputes, and multi-stage issues. Service quality dropped, and processing complaints ended up costing more money than was saved on salaries. Klarna became a corporate symbol of how "AI replacing human workers" ends up being more expensive than retaining the existing team.

Between these two poles lies your business

The space between these two poles is where CTOs, IT directors, HR directors, and general managers make operational decisions. At Anthropic, AI writes code, but the team is growing, because the focus is shifting to architecture, security, and product — areas that require human expertise. At Klarna, AI fully replaced employees, but failed to handle the actual work those employees were responsible for, and cost the company more than the payroll savings it generated.

The key question every manager should ask themselves in 2026 goes as follows: what exactly does the employee we plan to “replace with AI” do — routine processing of incoming workflows, or one-off complex decision-making? These are two distinct tasks. The first can be handed over to AI aggressively. The second cannot, at least with the current generation of models.

Russian Paradox

In Russia, the situation looks like almost a mirror image of that, which is strange. On the one hand — the number of IT vacancies fell by 41% over the course of a year, the HH index (number of resumes per vacancy) in the IT sector hit 22.9 in March 2026 — this is almost twice the threshold value at which the market is considered “extremely overheated” in favor of employers. More than half of laid-off IT specialists were made redundant, rather than leaving voluntarily.

On the other hand — demand for specialists with AI skills over the summer of 2025 grew by 66% year-over-year. It turns out that the market is contracting at the overall IT level, while expanding sharply at the AI skills level. This is not a paradox — this is role restructuring: positions with outdated skills are being phased out, and they are replaced by new ones where AI is an embedded tool, not a threat.

Role restructuring, not replacement

The English term that's currently making the rounds in AI analytics is rebundling. Ben Thompson of Stratechery coined it in the 2010s for the media industry: first, the internet broke newspapers apart into individual articles (unbundling), then reassembled them into new products with different features (rebundling). Today, the same process is happening with roles within companies. In Russian, this is referred to as role reassembly or function restructuring.

In layoff reports, replacement and reassembly look identical — "X people were cut". Operationally, however, these are two completely different scenarios. With replacement, we eliminate an entire function. With reassembly, we cut part of a function's responsibilities, regroup the remaining ones, add new ones — and end up with a company where a single person does the work that four people performed in the previous structure.

According to LinkedIn, the first two months of 2026 saw a 15% drop in new software engineer vacancies compared to 2025. At the same time, the same data shows that companies whose developers use AI tools produce 40–55% more code per sprint at comparable quality. A team of 10 developers using AI does the work of a 15-person team without it. This is reassembly in its purest form.

What AI can't do — and won't learn to do anytime soon

The phrase "AI will replace everyone tomorrow" usually conceals the idea that AGI is just around the corner. That is not true. Modern models face four hard barriers, and each one requires years of work to overcome. The first is data: all of humanity's core texts have already been fed to models, and synthetic data leads to degradation. The second is energy: current scaling rates are hitting physical limits; gas and coal cannot be extracted fast enough to keep up. The third is quantum computing for interconnected systems (weather, markets, cities) — the timeline for this is 10–15 years. The fourth is real-world sensors: AI exists only in the tiny, confined space of text and images, and it lacks the sensors needed to achieve true general intelligence.

This means there are categories of tasks that AI will not be able to handle in this cycle: complex multi-step customer conflicts (the Klarna case), architectural solutions with contradictory input requirements, accountability for business decisions, and emotional labor. Anyone who attempts to automate these tasks will repeat Klarna's mistake, only at a much higher cost.

Who to hire in 2026

If your company is hiring for a developer, analyst, marketer, lawyer, or HR role in 2026, I have one recommendation. Add a mandatory line to the job description stating "experience using LLM tools for work tasks", and conduct a technical screening with specific prompts. Skip the theoretical question "are you familiar with ChatGPT" — instead, assign a real task: have the candidate use Claude, GPT, or Gemini to produce a working deliverable aligned with the role's responsibilities in 20 minutes.

Based on our experience at Alpina Digital, the productivity difference between an AI-native employee and an employee unaccustomed to AI in the same role ranges from 2× to 4× — this is an operational metric, not a marketing figure. You can pay them the same salary, but complete tasks at drastically different speeds.

How to retrain an existing team without laying anyone off

The biggest strategic mistake of 2024–2025 has been mass layoffs "in the name of AI" with no attempt at retraining. This is exactly the "formula for failure" outlined in recent research from Stanford and McKinsey: companies placed their bets on the tool, overlooked employee training, failed to set up internal support systems — and ended up with costly reversals, just like Klarna.

At Alpina Digital we took a different path. Between 2024 and 2025, we ran the entire team through an AI skills intensive program — from copywriters and editors to backend developers. We rolled out corporate training formats, held internal demos, and shared successful use cases across the team. The result: the book release cycle at our publishing house was reduced by a factor of 4.5, with zero layoffs for "redundancy". No employees left the company — their roles and responsibilities simply shifted.

Let's assess the economics honestly

If you are a CFO or general director planning to make layoff decisions, I have three questions for you. First, have you calculated the cost of quality: not the operator's salary, but the real cost of an unresolved customer complaint, a lost contract, and reputational risk. Second, have you calculated the cost of recovery if the experiment fails: hiring, training, and ROI on pivoting, as Klarna experienced. Third, have you crunched the numbers for the alternative scenario: what happens if instead of cutting 700 operators, you retain 700 operators and add AI, and use this to grow your overall business volume rather than just boosting margins.

In most real-world cases we have encountered with our clients, the third scenario outperforms the first in terms of pure economic benefit. Klarna learned this the hard way, and far too late.

A counter-question for employers

When people ask me if AI will replace my developers, I always respond with a counterquestion: who do you plan to hire instead of them, and who will train those who stay? If the answer is "we won't hire anyone to replace them, and we won't train the remaining ones" — Klarna 2.0 is at your doorstep. If the answer is "we hire AI-native people and train the current team" — you are on the right track, and for you, the current 80,000 people laid off each quarter are not a threat, but an opportunity to poach strong AI-native talent from the market.

External measurements confirm this. According to Microsoft Work Trend Index 2025, 75% of knowledge workers already regularly use AI in their work, and among so-called "AI power users" 80% do things that were unavailable to them a year ago. The only question for an employer is whether they have a segment of employees within the company who fall into this category, and what you are doing to increase that number over the next 12 months.

Rifle vs. Spear

Let's wrap up with the core thesis that holds true both on an individual and corporate level. Your developers will not be replaced by AI. They will be replaced by other developers who work with AI. In this sense, the problem lies not with the employee, but with the employer: a company that has not established AI practices will lose out not to "armies of algorithms", but to a neighboring company where 200 employees get the work of 800 done using tools.

The timeline of this debate — between ultra-optimists (5 years until AGI) and cautious voices (20+ years) — puts you at the final stretch of your team's professional careers. The question is no longer whether this future will arrive, but who will be in the front car: you or your competitor. If you have to choose, I'd recommend being the first.

More case studies and guides on corporate AI implementation are published on our Telegram channel «It's All in the Prompt».

Sources

• Yahoo Finance / Tech Trenches — Dario Amodei at the Council on Foreign Relations (March 10, 2025): 90% of code in 3–6 months

• StartupNews.fyi — Mike Krieger at the Cisco AI Summit (February 2026): “effectively 100%” AI-generated code

• Tom's Hardware — 80,000 IT layoffs in Q1 2026, 47.9% due to AI (per Nikkei Asia)

• CBS News — Challenger Report: AI caused 26% of layoffs in April 2026

• Reworked — Klarna rehires people after AI replaced 700 operators

• CNews — Paradox of the Russian IT industry: 41% reduction in job vacancies amid AI skills shortage

• Microsoft Work Trend Index 2025 — 75% of knowledge workers use AI, “AI power user” profile

• Stratechery (Ben Thompson) — the concept of unbundling / rebundling

• Original webinar “Will artificial intelligence replace us?” (Zhemal Khamidun)