Will AI Eliminate Jobs, or Expose the Quality of Work?

Let’s call her Ayşe. The name is made up, but the job is familiar. She has worked on a customer support team for three years, and for the past six months she has been doing that job a little differently.

She used to write nearly every reply from scratch. Now the AI drafts a suggestion; Ayşe checks it, fixes it, and rejects it when she needs to. Simple requests close faster, while complex cases still land on her desk. Her manager says, “Productivity is up.” Ayşe says, “My job has changed, but it’s still my job.”

They’re both right. But here’s the point people miss: AI itself doesn’t decide whether this story ends well or badly. That depends on the data you put behind the model, how the process is designed, quality control, the customer experience you’re aiming for, and, above all, what management intends.

The debate about AI and jobs has long swung between two extremes. On one side: “Everything will be automated.” On the other, the reassurance that “nothing much will change, don’t exaggerate.” The data we have today supports neither. The more realistic picture is this: jobs don’t vanish overnight. They break apart first, then get redesigned. And if that redesign is done badly, both the project and the people get hurt.

What the big numbers say, and what they don’t

The ILO’s 2025 update says that one in four jobs worldwide could be transformed by generative AI. The key word is transformed, not erased.

The IMF points in a similar direction: about 40% of jobs will be affected by AI, rising to as much as 60% in advanced economies. The World Economic Forum’s Future of Jobs Report 2025 expects 170 million new jobs to be created by 2030 and 92 million to be displaced, a net gain of 78 million.

These numbers don’t say “relax, there’s no problem.” Quite the opposite: they say the transformation won’t be evenly distributed. For some it will bring new opportunities, for others job losses, and for most, a very different working day under the same job title.

Then there’s the ILO finding I consider the most important and the least discussed: according to its analysis, about 3.3% of global employment falls into the highest generative AI exposure category. That’s why it’s more accurate to read the issue not as “jobs will disappear entirely,” but as “the content of jobs and the way tasks are divided will change.” Set that number next to the “AI will take our jobs” headlines, and you can see how oversimplified the debate has become.

About 3.3% of global employment is in the highest generative AI exposure category. — ILO, Generative AI and Jobs: A 2025 Update

Field data: speed, yes; value, not always

Macro reports show the direction, but the real story plays out on the ground, and the findings there are more interesting.

In an NBER study of customer support teams, Brynjolfsson, Li, and Raymond found that AI assistance raised average productivity by 14%. More striking still, for less experienced workers the gain rose to 34%. In Noy and Zhang’s experiment published in Science, professional writing tasks took 40% less time and quality rose by 18%. GitHub’s Copilot research found developers completing certain tasks 55% faster.

So far, so good. AI can lift less experienced workers in particular, and it can turn the knowledge patterns of senior workers into something like a shared workspace. That is very valuable.

But there’s another side to the picture. Many AI projects never make it from pilot to production, or fail to deliver the expected impact when they do. As I discussed in the second article of this series, The GenAI Divide, a 2025 study attributed to MIT NANDA, found that despite $30–40 billion in enterprise generative AI investment, most pilots stalled or never translated into measurable profit-and-loss impact. That doesn’t mean “AI doesn’t work”; it’s a warning about poor data, unclear process ownership, and weak value measurement. Gartner wrote in 2026 that at least half of generative AI projects are abandoned after proof of concept, citing poor data quality, inadequate risk controls, rising costs, and unclear business value.

This is where an old but still very true principle comes back: garbage in, garbage out. Feed a system dirty, inconsistent, incomplete, or poorly contextualized data, and even the best model will hand you bad results dressed up in elegant sentences.

As of 2026, I believe the real dividing line in AI projects isn’t the choice of model; it’s the maturity of data and processes. When something that worked in the demo stumbles in production, it’s rarely because the model is “dumb.” It’s because the customer data is scattered, the product hierarchy is inconsistent, the process owner is unclear, the success metric is vague, and human approval was never designed. Then the whole thing gets labeled “AI,” and everyone expects a miracle.

The model doesn’t work miracles. It amplifies the process you already have. If the process is good, it speeds it up; if the process is bad, it scales the errors.

Klarna: not a U-turn, a recalibration

That’s why the Klarna story deserves a closer reading. At first glance it’s a very tempting narrative: “Klarna replaced people with AI, regretted it, and started hiring humans again.” That’s not entirely wrong, but it’s incomplete.

In early 2024, Klarna announced that its AI assistant was handling about two-thirds of customer service chats, the equivalent workload of 700 full-time employees. Statements reported by Reuters in 2024 described the company slowing hiring, shrinking through natural attrition, and cutting resolution times significantly thanks to AI.

Then, in 2025, the tone changed. In comments reported by Bloomberg and TechCrunch, CEO Sebastian Siemiatkowski said the company had focused too much on cutting costs, had underestimated the quality dimension of customer service, and that customers should be able to reach a real person when they need one. Klarna began rebuilding its human support, but framed this not as a one-to-one return to the old model; instead, it built a new, more flexible setup suited to remote work.

The key lesson is this: Klarna never said “AI didn’t work.” On the contrary, its 2025 SEC filings describe the AI assistant still handling most customer service chats, delivering substantial cost savings, and being widely used across the company. So the turnaround wasn’t about abandoning the technology; it was about redesigning the moments that call for quality, trust, and a human touch.

That’s exactly what makes Klarna interesting to me. AI carries the volume; people carry the trust. Simple, repetitive, measurable requests shift to the machine. In moments involving exceptions, disputes, fraud, dissatisfaction, or emotional tension, people move back to the center.

This lets us say something more mature about AI and work: AI can reduce some jobs, raise the quality of others, and, in some places, make human value more visible. Which of these happens depends on the design decisions made at the outset.

Three companies, three choices, three outcomes

Alongside Klarna, two more examples are instructive.

Verizon used a Google-based internal assistant to cut call times in customer service. According to Reuters, rather than shrinking its 28,000-strong service team, the company steered those employees toward more sales opportunities and saw sales rise by roughly 40%. Here, AI became a tool for growing revenue, not for cutting headcount.

Lyft, working with Anthropic, said it had significantly reduced average resolution times in customer support. But it made a point of stressing that people would continue to handle sensitive areas such as safety and fraud. In other words, it defined from the outset where it would hold the human line while improving efficiency.

Put these three examples side by side and the point becomes clear. With the same technology, you can cut costs, grow revenue, or redesign service quality. AI isn’t an automatic destiny; it’s a lever that accelerates management choices.

Where you place the lever matters. Put it in the wrong place and you won’t grow faster; you’ll break faster.

Work breaks apart before it changes

Here’s the subtlety that often gets missed: AI usually transforms work at the level of tasks, not roles. If a marketing manager’s job is a mix of strategy, content, analysis, reporting, and coordination, AI first speeds up content and reporting. The title stays the same, but the texture of the job changes.

When only 3.3% of global employment sits in the highest exposure category, asking “which professions will disappear?” is too blunt. The better question is: which tasks within this role will be automated, which will be strengthened by AI, and which have to stay with people?

The answer isn’t always technical. AI can write a standard report draft, but a person still knows the context in which it will be read, which political sensitivities it might touch, and how it should be explained to which client. A support bot can explain the returns procedure, but a person is better at sensing that an angry customer is really looking for trust, not procedure.

Garbage in, garbage out applies to workforce design as well. If your role definitions are fuzzy, your process owners unclear, your data dirty, and your only success metric is “how many minutes did we save?”, AI won’t give you a good transformation. It will give you an expensive illusion.

The real risk for junior roles

Then there’s the question of entry-level roles. I think this is where the most critical risk lies.

Routine writing, basic analysis, standard customer replies, simple pieces of code, and documentation used to be where junior employees learned the job. Now AI can do much of that very quickly. In the short term, that’s tempting for managers: “AI does this now; let’s not hire a junior.”

But that decision can carry a heavy cost five years down the line. Remove the junior role, and you weaken the pipeline that produces seniors. The field data argues against it, too: the biggest productivity jump in the NBER study, 34%, was measured precisely among the least experienced workers. Stanford HAI’s 2026 AI Index reports that employment of US software developers aged 22 to 25 has fallen by about 20% since 2024. That figure alone doesn’t explain the whole world, but it’s an important warning sign.

To me, the right response isn’t to treat that 20% decline as fate and eliminate the junior role; it’s to redesign the role. A junior employee should no longer be just “the person who does the simple work.” They should be the person who checks AI output, questions data quality, writes test scenarios, learns the domain, and understands the reasoning behind decisions.

In other words, the junior learning path is changing. But if that path closes completely, in a few years everyone will be asking, “Why can’t we find good seniors?”

The real question for Türkiye

In Türkiye, TurkStat’s 2025 AI statistics send an important signal. Generative AI use among individuals stands at 19.2%, rising to 39.4% among 16- to 24-year-olds and 30% among 25- to 34-year-olds. Among enterprises, the AI adoption rate is 7.5%.

What this gap means is simple: the generation entering the workforce sees AI not as an extra skill but as a natural way of working. Organizations, meanwhile, are still lagging on policy, data readiness, process ownership, and measurement.

So the real question for Türkiye isn’t “which jobs are at risk?” It’s this: will organizations be able to design this new way of working together with data quality, governance, and people development?

If they can’t, two things will happen. First, employees will keep using their own tools, and shadow AI will grow. Second, pilots that look good inside the organization but deliver no value in production will multiply. Data risk will grow, and a weary “we tried AI, it didn’t work” mood will set in.

Closing: not the model, but operational maturity

I started this series with a single question: is AI just making organizations faster, or is it also redefining who they are?

Four articles later, my answer is clearer. Ethics is the control architecture itself. In procurement, the real issue is reversibility. Data security is a design question that has to be considered together with employee behavior and tool choice. And the workforce isn’t just an automation debate; it’s a debate about data quality, process maturity, customer experience, and human development.

The Klarna case shows this well. AI carried the volume, but it couldn’t erase the value of human contact. The Gartner and MIT findings point to something else: AI projects built on bad data, unclear goals, and weak governance can’t deliver lasting value, even with powerful models.

So if I had to sum up this article in one sentence, it would be this: AI doesn’t fix a bad process; it just exposes it faster.

It won’t be AI alone that decides what Ayşe’s job looks like five years from now. That will be decided by how her company organizes its data, how it measures quality, where it places people, and how it shares the productivity gains.

The competition ahead won’t be among the companies that use AI the most, but among those that design AI, data, and people together best. The rest is detail.