In spring 2026 almost 4000 IT people in Ukraine were surveyed. The results are the kind that make you want to close the tab and pretend you never saw them.
44% of those looking for work voluntarily lowered their salary expectations. Among developers it is 46%. Almost every second programmer in the country walked into an interview and said it himself: "I am ready to work for less." Nobody asked him. He offered 🙂 HR, designers and QA dropped even harder, more than half of them cut their price. AI/ML engineers are the only ones who did not bend. Because there are still not enough of them to go around.
Frontend developer: 132 applications for one opening. A hundred and thirty two people fighting for a single seat. Median job search: 11 weeks. For two and a half months you send out resumes, sit through calls, do test tasks. Only to hear at the end "we picked another candidate." And senior devs, leads and managers who change jobs often lose salary. They do not grow. They fall 📉
The 50 largest IT companies in Ukraine laid off almost 11 000 specialists in a single year. Eleven thousand. EPAM, SoftServe, GlobalLogic, and down the list. IT export is at its lowest since the start of the war.
And against all of this, my interview. A Ukrainian product company, not a no-name. The call goes well. We discuss the stack, the tasks, the product roadmap. HR is smiling, everything is smooth. At the end, the classic: "What are your salary expectations?" I say: from $2500, but open to discussion, depends on the tasks, the scale, where the company is heading. A normal answer. An open position.
Silence. The smile is gone. "Thank you, we will call you back." You know that "we will call you back"? It is the Ukrainian IT equivalent of "go to hell, but politely" 😅 They did not call. Two days later the same opening. $800. Eight hundred dollars for a full-stack engineer. It is not even funny. It is a diagnosis.
That same week. A client from the EU. Wrote in and described the problem: "Our analysts spend a full day on each lease. We need this done in minutes." I answered with how I would solve it. He said: "Makes sense. When can you start?"
No test task. No "send us a PDF portfolio." No three rounds of interviews. A person described a pain, heard a solution, signed a contract. The amount: x8 of the same $2500 I was politely turned down for.
The difference is not the money. The Ukrainian market asks "how much does an hour cost?" The European one asks "can you solve the problem?" The first haggles over a resource. The second pays for a result.
But say they actually hired you
Fine. You agreed. $800, $1200, $1500, does not matter. You are inside. Working. Thinking: well okay, at least it is a product, at least it is interesting, at least it is experience. And then it starts.
You see a bug. Or a feature you could ship in half an hour. You sit down, open the editor. And then: stop. You need access. You have repo access, but not staging. You message the devops. Devops says: "Open a ticket." You open a ticket. The ticket needs the team lead to approve it. The lead is on calls until lunch. After lunch he looks, asks a question, wants you to clarify the scope. You clarify. He approves. Devops sees it the next day. Grants access. Two days just to start a 30-minute task.
And then the task itself. You did it. Needs code review. The reviewer is busy on another sprint. You wait. The reviewer looks, leaves three comments, one on point, two about style. You fix them. You wait for the second approval. Second approval from the lead. The lead is on a call again. Merge on Thursday. Deploy on Friday. A half-hour task took a week.
And it is like this every time. Every task grows a shell of approvals, access requests, tickets, standups, reports. You report on every micro step. Not because the task is hard. Because the process is built that way. Process for the sake of process. Bureaucracy for the feeling of control.
And you sit there and think: this week I did a day and a half of real work. The rest went into getting permission to work and proving that I work.
And now they want to shove AI into this machine
This machine that spends a week approving a 30-minute task. These people who are afraid to grant staging access without a ticket. This process with three approvals for one PR. Now tell them: "We need to integrate AI into the product."
Know what happens? Nothing. Absolutely nothing 🙃
93% of Ukrainian companies say they "use neural networks in business processes." Ninety three. Sounds like Ukraine is ahead of the planet. In reality it means someone in marketing generates texts through ChatGPT. That is it. There is your adoption.
Uklon, tens of millions of users, tried to launch an AI support bot. They built one "universal" bot. The bot started mixing up topics, losing context, talking nonsense. They had to rebuild it into a multi-agent architecture. This is 2026. What European business ran back in 2024, the largest Ukrainian ride-hailing app is only getting to now.
Vodafone Ukraine deploys AI solutions exclusively on their own GPUs, because they are scared to send data outside. Monobank refuses external AI providers entirely. This is not a security-first approach. It is paranoia that paralyzes. While you build your own GPU cluster so that god forbid a request leaves for an API, your competitor in Denmark has already automated half of their operations.
Denmark: 42% of businesses use AI in production. Finland: 38%. Eastern Europe: under 10%. That is not a gap. That is a chasm.
While the market haggles over a developer's hour, the developer got cheap
Here two things need to be connected, the ones usually discussed separately. The market that chokes a person with process and squeezes the price. And the tools that, in the same span, learned to do exactly what that person was paid for. One lands on top of the other. And you get a perfect storm.
The same code you pay developers for. That you haggle over in interviews. That you build processes, reviews and sprints around. That code is now written in seconds 🔥 Literally.
Claude Code, Cursor. You open it, describe what you need, get working code. A component? 15 seconds. An API endpoint? 20 seconds. A whole page with a form, validation and a server submit? A minute. A test for it? Another 30 seconds. What a junior spent half a day on. What a mid spent a couple of hours on. AI does it while you pour coffee.
And this is not a demo. Not "under ideal conditions." I work like this every day. Right now.
That EU client with his lease contracts, remember? Analysts spent a full working day parsing each contract. Reading, pulling out terms, checking risks, entering it into a system. One contract, one human day. I built a system that does it in minutes. Upload, parse, extract terms, flag the risks, a ready report. Building something like that used to take a team: a backend dev, a frontend dev, a data scientist, a devops, a lead. And a year and a half to a working version. I did it alone. In a couple of months. From architecture to production.
Not because I am a genius. Because the tools changed, and the market has not noticed yet.
"Come on, AI writes garbage code, you still need a senior to clean it up"
I hear this objection constantly. And it is correct. AI really does generate code you have to read, check and half rewrite. It lies with confidence, invents functions that do not exist, breaks on edge cases. All true.
Except the conclusion from this is the opposite of what the skeptics draw.
Yes, you need a person who cleans it up. But it is not the person who "also knows how to write code." It is the person who understands what the code should be. Who sees that the agent generated plausible nonsense, and knows why it is nonsense. The value used to be in writing the code. Now the value is in telling a correct solution from one that just looks correct. This does not cancel the developer. It kills the developer who could only type code from a spec, and raises the price of the one who understands the whole system.
AI did not devalue engineers. It devalued typing. And that, it turns out, was never the same thing as engineering.
So what do you pay for now
$200 a month for Claude Code. Works 24/7, does not get sick, does not burn out, does not ask for a raise. But it is not a magic button. Without a head that understands what to build, it is useless.
So what do you pay $800 for now? What do you pay $2500 for? What do you pay a developer for at all?
You pay for understanding. For knowing why this code is needed. What problem it solves. How it fits into the business process. What compliance requirements exist. What it integrates with. What breaks if you do it wrong.
AI does not know that an Austrian Werkvertrag has different liability terms than a Czech Smlouva o dílo. Does not know that myUSCIS has no public API. Does not know that D2C support in the EU needs a GDPR filter on every reply. It will write the code. But what to write, it has no idea.
Here is the core problem: the Ukrainian market spent 15 years teaching developers to close tickets, not to understand business. "Here is the spec, here is Jira, here is the deadline. Do it." And people learned. They close tickets brilliantly. Only tickets are now closed by AI 🤷
And the person who can only do that costs $800 and is falling. The person who understands the process and builds the whole system costs x20 and is growing. Because there are few of them. Because nobody taught this. Because the outsourcing mindset does not assume a developer should think. He should do.
The age of one person
One person in 2026 closes the full development cycle. Not just "frontend" and "backend." The full cycle. Here is what that means concretely.
System architecture. Stack choice. Interface design. Frontend: React, Next.js, responsive, animations, accessibility. API layer: endpoints, auth, validation, rate limiting. Backend logic: business rules, task queues, workers, webhooks, integrations with external services. Database: schema, migrations, indexes, RLS policies, backups. AI logic: prompts, tool-use, streaming, multi-agent pipelines, eval loops. Infrastructure: CI/CD, deploy, monitoring, alerts, logging. Security: KMS encryption at rest and in transit, RLS at the database level so one tenant cannot see another tenant's data, scanning incoming files before processing, rate limiting at every layer (edge, API, AI tokens), WORM storage with a hash chain on every action of the system, crypto-shred for GDPR right to erasure. Documentation, testing, iterations with the client.
This is not fantasy. This is my normal workflow. Every project I shipped in 2026 went through all of these stages.
This used to require a team. A frontend dev, a backend dev, a devops, a designer, a QA, a lead, a project manager. Five, seven people minimum. A year, a year and a half to a working product. Now one person with the right tools and domain understanding assembles the same thing in two or three months 🚀
Not because he works 20 hours a day. Because AI took the routine, and he does what AI cannot: he thinks. Digs into the client's business. Makes decisions. Takes responsibility for the result.
The Ukrainian market, meanwhile, hires 15 people to spend two sprints discussing the architecture and another two sprints approving it.
Your "get into IT" courses are training dead people
In parallel with all of this, a whole industry in Ukraine takes money from people to teach a profession that no longer exists.
"Get into IT in 6 months." "Become a frontend developer." "Employment guaranteed." Thousands of people pay $500-1500 for courses that teach them to build landing pages and close tickets from a spec. They are promised a $1000-1500 salary and a "stable career."
Reality: junior and trainee specialists are 12.6% of all IT people in the country. Requests for junior positions come in less and less. Monobank does not plan to hire juniors at all. Most junior openings in 2026 already require commercial experience. Commercial. From a junior. The same one who just finished a course 🤡
A landing page, AI generates in 2 minutes. A test task, AI solves better than a course graduate. A person paid $1500 for training and walks into a market where his skills are worth less than a subscription to the tool that does the same thing.
These courses prepare people for 2020. It is 2026. The difference is not six years, it is a whole era.
HR does not understand who they are looking for
Fine, courses are courses. But now the employer side.
A typical opening. "Fullstack Developer." Requirements: React, Next.js, Node.js, TypeScript, AWS, Docker, CI/CD, PostgreSQL, Redis, AI experience a plus, English Upper-Intermediate. 3+ years experience.
This is a description of a $4000-5000 specialist minimum. A person who knows all of this at a working level costs money. This is not a junior after a course.
Budget: $1200 🫡
HR sees no contradiction. The requirements were copied from last year's posting. The budget was approved by a finance director who thinks "a developer is a developer." The applicants are juniors who wrote React on their resume after a three-month course. The product does not move. The company complains: "there are no good people."
The people exist. That same specialist on the foreign market gets 40-80% more for the same amount of work. Just a remote contract with a Western client. No relocation. No visa. Same chair, same laptop. A different Slack.
Why would he want your $1200 job?
Seniors leave quietly
And this is probably the scariest part. Not the layoffs. Not the export drop. Not AI. The best simply vanish quietly.
No farewell posts on LinkedIn. No forum drama. No haggling over salary. At some point a person takes one foreign client. Then a second. Then a third. And that is it. He is still in Ukraine. Still writing code. But the Ukrainian market no longer sees him 👻
58 000 Ukrainian developers already work abroad. That is the official number. And how many sit in Kyiv, Lviv, Kharkiv quietly building for clients in Berlin and Amsterdam, nobody counted. Because formally they never left. They just stopped being available.
There is a job posting that hangs for months. There is an HR who complains. There is a product that stalls. And the people who could pull it all out are working on a different market. For different money. At a different pace. And there is no reason for them to come back.
What comes next
By analyst forecasts, by 2027 80% of engineers will have to retrain for AI-assisted development. Eighty percent. Not "nice to learn." Have to. Otherwise irrelevant.
The number of jobs for developers aged 22-25 has already dropped 20% since generative AI appeared. Twenty percent. Juniors just entering the profession are walking into a door that is closing. At the same time demand for developers over 30 grew 34%. The market does not need hands. The market needs thinking heads.
Large Western companies with 10 000+ developers are already building internal AI platforms with one goal: remove outsourcing. Not reduce. Remove entirely. Replace it with efficiency from AI tools and targeted hiring of people who know how to work with them.
Specialists with AI skills earn 56% more. Not 5%. Not 10%. Fifty six percent more than colleagues without those skills. On the same market, in the same companies.
And here is the main problem. Not the money, not AI, not outsourcing. The problem is speed.
Technology now changes faster than people can adapt. This is not a figure of speech. Literally. You learned a framework, and in half a year it is outdated. You mastered a tool, and in three months a version shipped that works differently. You built a process in a team, and in a quarter that process can be replaced by a single prompt.
Technology used to change every 3-5 years. There was time to adjust. Now in one year as much happens as used to happen in ten. And the gap between those who run and those who stand still grows non-linearly. It doubles every few months.
The market cannot keep up. HR cannot. Developers cannot. Companies cannot. Education cannot. Everyone is chasing a train that already left. And the next one will be faster still 🚄
And here is where people usually wait for the takeaway
The advice. "Learn AI, master the tools, everything will be fine." It will not.
Most people will close this tab and in fifteen minutes forget 80% of what they read. Not because the text is bad. Because changing what is familiar is scarier than enduring what is bad. Easier to sit on $1200 and blame the market than to learn a new domain from scratch and go where they pay for results, not hours. It is always easier to wait.
But "for now" is the most expensive word in IT right now. "For now there is a project." "For now they pay." "For now AI has not learned to do my job." Every "for now" has a shelf life. And it shrinks with every model update.
I am not writing this to scare anyone. I am writing this because I stand on this side of the gap myself. A year ago I was politely turned down for $2500. Now I do what used to require a team of seven, and EU clients do not ask "how much does an hour cost." They ask "when can you start."
The difference between me and that frontend dev with 132 competitors for one opening is not talent. It is one decision: stop selling hands and start selling understanding.
The question that decides everything is simple. Do you understand why you write code, or do you just write code? If the latter, bad news: you are not competing with an offshore dev for $15 an hour. You are competing with a machine for $0. If the former, if you see the whole system and can turn someone else's pain into a working product, then 2027 will be the best year of your career. No joke. Because there are only a handful of such people, and demand for them only grows.
The window is still open. But it is closing. And with every model update, faster 🚪
By the way, I am not theorizing. While the market talked, I built. Fresh 2026 projects, put together solo from architecture to deploy, are here: khabenko.com/portfolio ✌️