AI is accelerating software development, but who will gain the experience needed to oversee the systems of tomorrow?

In November 2000, mBank launched as Poland's first fully online bank. It marked a turning point in the development of Polish digital banking. Within five years, the number of people using online banking services had grown from approximately 50,000 to nearly 3.7 million.

Behind this transformation were people: engineers, developers, and architects who built solutions at a time when the market was only just taking shape. Many gained their experience working directly on pioneering projects, learning new technologies on the job, alongside more experienced colleagues, and discovering the real-world consequences of their decisions. Today, they are architects, technology leaders, and executives whose expertise underpins a significant part of Poland's IT industry.

I bring this up because the career path that allowed them to gain that experience is becoming increasingly difficult for the next generation of engineers to access in 2026.

Numbers Every CIO Should Pay Attention To

According to the H1 2026 Labour Market Barometer published by Just Join IT and RocketJobs, junior positions account for approximately 5% of IT job postings, compared with around 56% for senior roles. Back in 2021 and 2022, juniors accounted for as much as one-fifth of the market. At major technology companies covered by SignalFire's research, hiring of recent graduates has fallen by approximately 65% compared with 2019.

Some of the most telling figures come from AI developers themselves. At Anthropic, the company behind the Claude model, the share of junior employees among newly hired technical staff fell from 27% to 9% in just one year. Across its entire technical workforce, the figure dropped from 21% to 14%. There is one important detail often missing from the headlines: the company tripled its workforce during that period. It did not stop hiring. It changed its recruitment structure, significantly reducing the share of junior positions among new technical hires.

Several factors contribute to this decline, including more cautious budgets and a correction following the pandemic-era hiring boom. However, it is becoming increasingly difficult to ignore the impact of artificial intelligence. Tasks that almost every developer started with, such as writing simple functions, tests, fixes, and documentation, are now largely handled by AI. According to the Dev Barometer survey, 42% of developers use AI to generate at least half of their code, up from 12% a year earlier. They also report saving approximately 13 hours of programming time per week. From the perspective of an individual company, the calculation seems straightforward: why hire a junior developer when an AI agent can do the job faster?

In the Just Join IT and RocketJobs survey, 69.3% of young job seekers said they had encountered job postings labelled as junior positions that, in practice, required mid-level or senior-level skills. Employers are increasingly looking for junior developers who are no longer really juniors.

Where the Equation Doesn't Add Up

Code written by AI does not exist in a vacuum. It needs to be reviewed, tested, integrated into an architecture, secured, and maintained. At Anthropic, researchers estimated that simply reading everything produced by AI agents would take even non-technical employees as much as half a working week. Engineers increasingly describe their role as overseeing multiple tasks running in parallel, while the median time employees spend in meetings increased by 20% over the course of a year. AI has not eliminated project costs. It has shifted them from writing code to verification, decision-making, and accountability.

I see this in conversations with clients, too. I am increasingly less likely to hear the question of how many people it takes to build a system. Instead, clients want to know whether someone on the team will understand what the system actually does once it goes live, connects to solutions built fifteen years ago, and encounters its first unexpected situation.

That kind of assessment requires something that cannot simply be generated: experience. To determine whether a solution proposed by a model will bring a system down under heavy load, introduce a security vulnerability, or break an integration, you need to have seen these things fail firsthand. This is why AI has, paradoxically, increased the value of senior engineers.

A Lesson from the Cockpit

Aviation went through a similar transformation much earlier. Automation took over a significant share of routine cockpit tasks, yet no sensible person suggests putting inexperienced people with no flight hours at the controls. The more automation takes over, the more important a pilot's experience becomes, because their role increasingly centres on situations where automation fails or circumstances fall outside the expected scenario. Airlines cannot, however, hire only captains with twenty years of experience. They need first officers who are still building their flight hours. Otherwise, in a dozen years, there will be no one left to occupy the captain's seat.

In 2026, the IT industry is behaving a little as if it wants to hire nothing but captains.

There is also a risk that is less visible in the statistics. We could end up raising a generation of engineers who are excellent at prompting AI models but have never had to independently diagnose a bug, design an architecture, or restore a system after an outage. I am not arguing that every developer must write everything by hand. I am arguing that you cannot effectively oversee technology without understanding its limitations. And that is not something you learn by looking at correct answers. You learn it when something breaks and you have to figure out why.

A Risk, Not a Foregone Conclusion

Every senior engineer was once a junior. If the industry continues to limit entry-level hiring over the long term, we could face a shortage of people capable of overseeing increasingly powerful AI systems in eight or ten years. An individual company may save money in the short term, but the market as a whole risks consuming the seed corn it needs for the future.

This is particularly important for Poland. For years, the country has attracted global companies by combining competitive costs with strong engineering capabilities. Today, however, quality, experience, and the ability to deliver complex projects are becoming essential to maintaining that position.

There is another possible scenario. AI could help young engineers develop their skills faster than previous generations. Instead of spending months on repetitive tasks, they could start participating in solution design and technical decision-making much earlier in their careers. But this scenario will not happen automatically. It requires access to real projects, responsibility for outcomes, and an experienced person working alongside them. Access to an increasingly capable AI model is not the same as gaining experience.

What Can We Do About It?

I do not believe in appeals for companies to hire juniors out of a sense of obligation. Instead, I see three approaches that make business sense.

The first is redefining the junior role. Rather than writing code from scratch, entry-level engineers can learn what is becoming most valuable from day one: reviewing, testing, and understanding AI-generated solutions under the guidance of someone who knows what to look for.

The second is building teams deliberately. Not every company needs to employ an architect, DevOps expert, or data engineer full-time. An experienced specialist or an entire team can be brought in for a specific project phase through outsourcing or body leasing. However, a well-designed engagement should leave behind more than just a working system. It should also leave the client's team stronger than before. At Sorigo, a company whose motto is "experience matters", we are seeing clients increasingly ask not just for a specialist, but for someone who can raise the capabilities of the entire team.

The third is measuring the impact of AI honestly. Not just the number of lines of code generated or hours saved, but also the time spent on verification, fixes, security, and maintenance. Only then can we determine whether automation is actually reducing the total cost of software development or simply shifting it onto more experienced and more expensive specialists.

The engineers who built Poland's online banking infrastructure twenty-five years ago were given their opportunity because someone decided it was worth investing in their development. Today, it is our turn to decide whether the next generation will get the same opportunity. Experience matters, but it does not come from nowhere.

Sources

mBank, One Million Customers for Its 5th Anniversary: https://pl.media.mbank.pl/47125-milion-klientow-na-5-urodziny

Anthropic, How Employment Is Changing at Anthropic: https://www.anthropic.com/research/employment-at-anthropic

Just Join IT and RocketJobs, Labour Market Barometer H1 2026: https://rocketjobs.com/barometr-rynku-pracy-H12026/raport

Just Join IT and RocketJobs, Junior Talent in the 2026 Job Market: https://rocketjobs.com/raport-junior-2026/raport

SignalFire, State of Talent Report 2026

BairesDev, Dev Barometer