Readme for programmers. Are these numbers true?

His company, Jassy promised, is becoming faster, more customer oriented and, especially, leaner. "We will need fewer people doing some of the jobs that are being done today, and more people doing other types of jobs," the CEO predicted. Overall, Jassy wrote, Amazon’s workforce will shrink in the coming years, a product of improved efficiency stemming from the extensive deployment of AI.

Workers, Jassy recommended, should see this as an opportunity to continue learning and developing, to keep up and improve. "Be curious about AI. There's so much more to come with generative AI. Andy.”

Humans are on their way out. Supercomputers are on their way in. Labor performed by people is to be replaced by, or at least combined with, AI. In factories, warehouses and fast-food restaurants. In offices and law firms. Perhaps, at some point, even in the fields, on the high seas and in the air.

Automation has been with us since even before the invention of the loom, which many historians see as the beginning of the industrial age. As time has passed, humans have delegated more and more unwanted tasks to machines. Real people, though, have always been responsible for thinking and planning.

That, though, is now changing.

Thus far, AI has primarily been known as a generator of text, sound and images, but the machines are now learning to act autonomously. So-called AI agents are programming software on behalf of their human masters or filling closets and refrigerators all on their own. They are running factories and designing automobile parts. AI bots remain patient when exasperated travelers want to rebook airline tickets. They can negotiate the price of potatoes for food companies all on their own, with no human involvement. In Albania, AI even has a seat at the cabinet table: In the future, a chatbot named Diella is to make decisions on the awarding of government contracts as a means of combatting corruption in the country. An actress famous in the country even lent her appearance to the robot.

And that, both experts and tech giants agree, is just the beginning.

According to the International Monetary Fund (IMF), almost 40 percent of jobs around the world are "highly exposed to AI,” meaning they could be performed by or with AI. In industrialized countries like Germany, that number rises to 60 percent because of the more advanced job structure. Around half of these jobs, according to the IMF report, could fall victim to AI: desk jobs, cashier positions and mail delivery, for example. But it is also true that other jobs will become more valuable through the implementation of technology: personnel managers or team leaders, for example, who will be able to more efficiently deploy their employees. Psychologists as well.

By as early as 2030, says the consultancy McKinsey, almost a third of all work hours worldwide could be performed by generative AI, which is capable of original creation and decision-making.

Silicon Valley believes the global economy is on the cusp of a new era, with some comparing the generation of artificial intelligence to the invention of the steam engine or electricity. That generation, the promises hold, will produce enormous gains in productivity, unimaginable scientific breakthroughs and the end of "3D” jobs, shorthand for "dirty, dangerous and demeaning.”

Economists are divided in their analyses. Some experts are warning of an enormous wave of job losses – while others believe a new age of 15-hour work weeks with full pay is dawning. The end of dependent wage labor is upon us, they say. It is very possible that the world is facing massive unemployment and empty state coffers. But it is also possible that the opposite will happen, that countries like Germany, long seen as ponderous and bureaucratic, will benefit from the transformation.

What, then, will AI bring? Boom or doom on the labor market? The welfare line or the hammock?

In world religions like Christianity and Islam, work is considered godly. Karl Marx also saw work as the meaning and essence of human existence, writing in 1853: "Society does not find its equilibrium until it revolves around the sun of labor.”

In 1983, the German magazine Psychologie Heute (Psychology Today) put a surfing Marx on the cover with the headline: "Less Work for All!” The computer age had just begun, and the story focused on what people might be able to do with all the spare time they would now have. Today, nobody would deny that personal computers, the internet and smartphones have made many things far easier and contributed to enormous prosperity. Such tools also made a globally networked economy and society possible in the first place.

But it is also true that many people had to undergo significant retraining, and this posed significant challenges for those who found such reorientation difficult. It is also true that we did not run out of work to do.

Might AI also end up being a tool enabling us to work more productively, think more broadly and become more inventive? Or will it simply make human labor redundant?

What Is an Agent? What Can They Do?

Silicon Valley is convinced that this time, they really have come up with something amazing. Previous "breakthroughs” like the metaverse proved to be little more than digital wastepaper, never finding a real application in real life. But "agentic AI,” crow the tech bros, is different.

And it’s true: The difference between AI as we currently know it and the solutions now being developed in the tech labs is roughly as great as that between a model toy and a self-driving car.

The stakes in this revolution are on full display in an industrial complex not far from the San Francisco International Airport. On the third floor of a glass building, Emily Glassberg Sands welcomes visitors into a room Stripe calls the "maker’s space.” It is filled with wooden tables, barstools, colored pens and rolls of colored paper, much like the creative space at a budding startup. Stripe, though, is not that. Payments worth a total of $1.4 trillion are processed each year by the company and the equivalent of 1.3 percent of the global GDP meanders its way digitally through the firm’s books.

The payment service provider is involved when a mother from Leipzig plans the next family trip through booking.com or when a smalltown kiosk in rural Germany orders more chips and cola from a wholesaler. The technology is used by Amazon and Google just as it is by Peloton. Such payments produce terabytes of data, which Glassberg Sands, who is responsible for AI at Stripe, is constantly analyzing.

"Classic artificial intelligence,” which calculates probabilities using algorithms and patterns, is being used less and less, says the executive. Instead, she has observed that banks or vendors, for example, now turn over the processing of customer payment disputes almost completely to digital agents. It used to be "an insane amount of paperwork,” says Glassberg Sands.

These days, an AI agent takes care of such cases. They examine the claim, collect all the relevant information and automatically submit all of it. The result: A 13 percent increase in claims won and millions of dollars saved. "Without a human in the loop,” she says.

For normal consumers, "agentic shopping” is industry’s great hope. The idea is that customers will simply hand over their credit card information to an AI agent, which will analyze the content of the closet or the refrigerator and automatically take care of purchasing another blouse and more beer.

Soon, Glassberg Sands says, there will be two versions of the internet. One for people, and one in which AI agents only interact with other AI agents. That, she says, will give us all "a completely new direction.”

The digital team included an artificial immunologist and a bioinformatics bot. According to (human) study leader John Pak, a "scientific critic” played a "quite essential” role – an agent whose skepticism kept the others under control.

The bots began interacting, both in inhumanly fast one-on-ones and in group meetings. "In the time that I had my coffee, they already had hundreds of research discussions,” says co-author James Zou.

Within just a few days, the agents developed a total of 92 so-called nanobodies, tiny proteins that work like antibodies, but are less complex and thus easier to calculate. In subsequent laboratory testing, two of them proved so promising that they are now being further examined by human scientists. It is, says Zou, nothing less than "a new paradigm.”

Agents are the topic of the hour on the top floors of blue-chip companies in Germany as well. Deutsche Telekom customers have long been less than impressed with the frequently headstrong AI chatbot called "Ask Magenta,” while employees use "AskT” to access the company’s amassed knowledge. Only now, though, is the technology really hitting its stride.

Together with Google, the Magenta developers have come up with a multi-agent system for monitoring the entirety of the mobile network. Among its tasks is automatically preventing reception problems or even outages from occurring at events and large gatherings. To do so, the AI system combs through the internet, including social media channels and traffic news, completely on its own. In a recent run, the agent identified almost 40,000 concerts, trade fairs, festivals, parades, flea markets and other events scheduled through the end of the year in Germany.

It then determines the state of the network at each event site. In some cases, the AI agent is able to optimize the network remotely, or it might recommend that additional mobile cell towers be set up, says new technical director Abdurazak Mudesir. AI agents have now become compatible with human "skill profiles,” that are specialized for specific tasks, says Mudesir. Here, too, there is a need for a concrete job description, a focused process for integration into workflows, regular performance reviews and clear guidelines on what internal data they are allowed to access. His strategy, says Mudesir, is to "encourage as many employees as possible to propose and develop specific AI agents.”

Not all Telekom employees are likely to be eager to participate. Additional productivity advances combined with even more automation are likely to have personnel consequences at some point. The last wave of job cuts at Telekom’s IT subsidiary in 2023 and 2024, which led to the elimination of about a quarter of the staff, was explained by company head Tim Höttges as being a consequence of AI.

Höttges has been open about his belief that taxpayers should pay the societal costs of his AI offensive. For the last decade, the Telekom CEO has been promoting the concept of universal basic income as a response to digitalization and automation.

It is suspiciously reminiscent of the state subsidized early retirement program that the company used to get rid of their aging, unneeded employees. Sponsored by society at large.

The Job Productivity Paradox

In almost all large companies, executives are seeing an opportunity to slim down, become more efficient and boost profits. It’s almost like a contest is underway: Who can offer greater growth rates? Who will cut more jobs?

According to the World Economic Forum, 40 percent of all companies are planning on reducing staff with the help of AI. Jim Farley, head of the U.S. automaker Ford, believes that agentic AI will replace around half of all white-collar jobs in the coming years – those challenging desk jobs that have long been considered safe.

Marianne Lake, head of Consumer & Community Banking at JPMorganChase, is planning with 10 percent fewer employees through the extensive deployment of AI technology, from customer service to the management of complex investment vehicles. Dario Amodei, CEO of OpenAI competitor Anthropic, turned heads with his recent prediction that the U.S. could soon be facing 10 to 20 percent unemployment. Within five years, Amodei said, "AI will eliminate half of all entry-level white-collar jobs.” Elon Musk, meanwhile, never one to shy away from hyperbole, predicted last year the end of paid work entirely.

Some of these predictions, to be sure, have more to do with propping up stock prices than they do with future realities. But those who have thus far viewed AI as yet another crazy idea from detached Silicon Valley billionaires are also likely well off course.

In the United States alone, companies invested almost $110 billion in the expansion of the technology last year. Expensive social and climate goals have had to make way for a prophecy that can only fulfil itself: Companies are investing in AI because their competitors are doing so. But they will only be able to win the race if they are able to quickly show falling costs. The easiest way for companies to do that is through job cuts – particularly in the U.S., where employees enjoy very few protections.

The revolution is devouring its own children.

None of this is particularly sustainable. Because even as the billions in costs and the job reduction plans are real, not much is happening on the other side of the purported equation. AI-fueled productivity boosts have only actually been realized by isolated companies in Europe and the U.S. Not, however, across entire industries or economies.

Which is why Amazon is trying to find the "sweet spot.” The point at which the efficiency curve of the machines intersects with that of human employees. "So that everyone does what they are best at,” says Amazon executive Parness. That is the sign of a "truly productive company.”

Because not everything that AI could theoretically take over actually makes sense in practice. In the Shreveport warehouse, for example, Sparrow, Robin, Cardinal and Proteus take care of a lot of the grunt work. But there is still room for humans. Amazon employee Chestney Flemming is standing in front of a screen showing a conveyor belt at one of the state-of-the-art sorting facilities. In front of him, Sequoia is selecting a blue box from a stack that has been brought to him by a different robot. Sequoia grabs the bin and sends it along the conveyor belt to Flemming, who then picks out the correct article to fill the customer’s order and returns the bin to the robot, which places it back in the stack.

It would perhaps be possible to deploy a humanoid robot to do Flemming’s job, Parness says, but it would hardly be efficient. A person needs just a few seconds to find the correct article in the box, pick it out and package it. A machine would take quite a bit longer. Humans are able to easily move among numerous different environments during the workday, fulfilling several tasks at the same time and have an "understanding of the world and each other.”

Machines have none of that.

Parness says he has a house and children, which is why he has a human housekeeper. She is able to see dirt and dust, can spontaneously respond to different kitchen appliances and understands when she should refrain from entering a room because someone is talking on the phone. "Of course I could buy an AI robot for several thousand dollars to take care of the cleaning,” says Parness. But it would have to be maintained and updated and would also need replacement parts. "Would it be worth it? Never.”

The same holds true for companies. It may sound like the end of work, but it’s not – not for everybody. For now, at least.

The German Dilemma

It’s all a question of pressure, seems to be the message from Alexander Wallner. Thus far, says the Germany country leader for the U.S. software giant Salesforce, AI hasn’t been a real "business case” for companies here. Chatbots were able to help simplify processes, he allows, but their answers weren’t always accurate and didn’t deliver any productivity improvements to speak of. Humanoid-acting agents, on the other hand, are "a once-in-a-century opportunity for many to return to the path of growth.” For the struggling auto industry, for example, or the retail sector. "The higher the economic pressure, the more repetitive the task, the more companies are investing in AI.” And not with the goal of getting rid of personnel. "Instead, they are snapping up those who are becoming available and thus covering their needs for skilled workers.”

Wallner’s comments are self-serving: Salesforce has strongly linked its fate to AI agents and is the global leader in programs that help companies manage customer relations, from call centers and chats with customer assistance to the development of offers. More and more processes are being automated, says Wallner. AI agents process fully 85 percent of support requests to Salesforce without assistance.

In the United States, demand is huge, he says, with companies there eager for "quick wins.” In Germany, Wallner says, the reflex is to first "plan and see where it can be efficiently deployed.” That sacrifices speed, particularly since the labor unions are extremely skeptical of the technology. Ultimately, though, he says: "Those who work with AI will replace those workers who reject AI.” Retraining at the age of 50 or 55 is becoming the norm, he says, and one’s own "adaptability” is decisive when it comes the question of: "Are you benefiting from this revolution or not?”

Which sounds a lot like flexibility and a new start – not exactly qualities that Germany is known for.

And yet, things could still turn out just fine for German workers. That, at least, is the belief of Anton Krüger, director of the German Research Center for Artificial Intelligence. Activities "for which you don’t need any common sense” can already be performed today by AI. "But you need common sense for an astounding number of activities. At least in Germany.”

In the U.S., says the professor, the labor market is quite different, with some jobs being so "fragmented and lacking in complexity” that AI could actually replace many tasks. In the long term, though, he believes that Germany has good cards. Here, says Krüger, the economy is organized such that more "everyday intelligence” is required. Employees "jump” between tasks, workflows are "more varied” and positions "more challenging.” Completely replacing that "with hordes of AI agents” would be considerably more difficult.

Thus far, Germany has profited in the long term from every wave of automation. Studies show that in the best case, AI could increase the country’s gross value added by 330 billion per year. Already, current levels of prosperity and welfare support can only be maintained because the degree of automation in the German economy has been continuously increasing for the last 25 years. To a level that is much higher than in the U.S., where a surprisingly high number of repetitive tasks are still performed by human assistants and interns.

Krüger says that Germany must, of course, pick up speed when it comes to digitalization and become more flexible in vocational training and continuing education. But the conditions for a "hybrid future for work” are "absolutely present.”

Thus far, though, implementation has been lacking.

Julian Wiedenhaus and his Hamburg startup Plancraft have turned Germany’s lack of digital diligence into a business: For the development of its AI agent, they went looking for an industry with a low level of automation – and found it in the good old trade crafts. Starting with a software program encompassing the business processes of roofing companies, plumbers and co., Wiedenhaus developed a "companion for the day-to-day of trade craft.”

A painter, for example, must only enter the measurements of the construction site in question verbally and the AI agent will produce an estimate and send it to the customer. Shortly before the end of the day, it will remind workers of the need to finish their daily log, which they then must only dictate. The agent takes care of paperwork, the office and personal assistant duties. Which means that the workers can focus entirely on what they trained for.

Wiedenhaus says that the product is already being used by around 20,000 customers. He has collected more than 50 million euros from investors and is currently expanding to Spain, Poland and the Netherlands. On average, according to customer feedback, the AI agent saves each worker eight hours per week, says Wiedenhaus. An entire workday that can now be spent on the job site.

Will it be enough to solve the skilled worker shortage? Does the labor market have sunny days ahead?

The Crisis for Programmers

No, it does not. Even if AI might help established, experienced employees become better and more efficient than ever before, the technology could still prove to be a job killer.

That, at least, is the message relayed by Thomas Dohmke in the city of Bellevue, just east of Seattle in the northwestern corner of the U.S. When the Berlin native took over the top post at Github in 2021, the Microsoft-owned programming platform was really only known to computer nerds. Today, it is used by almost 150 million people, who use Dohmke’s tools to write computer games and websites, mobile apps and PC software.

These days, almost two-thirds of all code comes from artificial intelligence. "What once felt like magic has become completely normal today,” says Dohmke in his spacious office with a view of Mt. Rainier out his office window, a snowcapped peak – even though it’s August – of 4,392 meters.

AI, says the Github CEO, is like the apocryphal iceberg, of which only the tip can be seen, with 90 percent lying under water. The world is "finally” realizing the full implications of the technology: Nobody has to learn a programming language anymore. "The idea is enough. The rest is taken over by our co-pilot.”

It is something of a paradox: There has never before been so much computer code in the world. Never before has so much software been needed, never before so much developed. And yet, in the last two years, more than a quarter of all programming jobs in the United States have vanished. According to government data, there are now fewer professional coders in the U.S. than there were in 1980, with the tech sector having eliminated 100,000 jobs this year alone. It is the largest wave of downsizing the industry has ever seen.

There will continue to be a need for programmers, of course, especially so-called software architects, says Dohmke – those well-trained, highly skilled individuals who design the increasingly complex structures of programs. But the job of "junior coder” will "change dramatically.” AI, he says, has become so present and so powerful that human programmers need a completely new skill set to keep up.

And that, to continue Dohmke’s metaphor, is likely just the tip of the iceberg. In the United States, a third fewer entry-level positions have been advertised since just 2023.

That development, of course, offers great opportunities to startup founders. Business ideas centered on AI agents are now attracting record amounts of venture capital and other investors, even for those without a vast network and also far away from well-established startup ecosystems. Droidrun from Osnabrück is one such example, a company launched by 26-year-old Christian Ninstel that produces AI agents for smartphones, enable users to control apps verbally. These agents extract textual information from an app and the user can then interact with the app via text or speech – to plan a route, for example, or to have messages read out. In Stockholm, meanwhile, 35-year-old Anton Osika’s company Loveable has become the fastest-growing AI startup in Europe, allowing people to write their own software despite knowing nothing about programming.

Not every school graduate is an entrepreneur, however. Plus, it is almost a sure thing that there will be a wave of consolidations among AI startups sooner or later as well.

As such, the only sensible plan, say scholars, is to approach the problem head on: Employees must become more flexible, universities have to rewrite their curricula and students must think hard about their choice of subject. Previous revolutions, such as the introduction of the personal computer, ultimately led to robust increases in employment and real wages, say labor market experts.

What is clear, though, is that some on the job market will fall victim to AI. And that the biggest victims will be those with lower levels of education or those who cannot or will not change. As such, AI will also be a social welfare issue.

What role, though, will humans continue to play in a work world infused with technological optimization?

Humans or Machines?

In the rainy Medienhafen quarter of Düsseldorf on a Wednesday in September, Rocco Bräuniger, the head of Amazon Germany, lays a communiqué from company headquarters in Seattle on the table. In it, Amazon executives lay out the next step in AI development: an autonomous agent able to take care of almost everything for the many millions of vendors on the site. "From ordering products to warehousing and pricing to the serving of ads,” says Bräuniger. "With just a couple of clicks, these AI tools will take care of it on their own.”

All those selling products via Amazon here in Germany, he says, will soon be able to use the technology. Free of charge. Which translates to more people than every before being able to offer their products through Amazon, build up a business, earn extra income, Bräuniger says. Optimize their product range. Increase sales. And "focus more on new products without the annoying administrative hassle.” All of it, says Bräuniger, will become easier than ever thanks to AI.

It could, in fact, be a gigantic step forward if technology were to take on those numerous boring and monotonous tasks. And if you believe Reid Hoffman, such a reality might not be all that far away.

As a multi-billionaire, there isn’t much he can’t afford – including a disheveled beard, crooked glasses and uncombed hair. Hoffman doesn’t need to impress anyone. The American founded the professional networking site LinkedIn and brought it to the stock market before then developing PayPal with Peter Thiel and, more recently, helping AI companies like OpenAI and Inflection out of their infancy. During his long Silicon Valley career, he has earned $2.5 billion. And he now spends his time writing engaging books about the future of humanity and the tech industry. His newest publication is called "Superagency,” about the wonderful new world of work.

Sitting in his home near Seattle before an indefinable painting – three dark brushstrokes on a beige background – Hoffman smiles as he says he doesn’t want to sound overly optimistic or even naïve. Work and the labor market will change drastically through the implementation of AI, he allows, and some of these transformations have already begun. There will be, he says, companies that blindly slash jobs.

But none of that will last, Hoffman believes. Companies are still in competition with each other, vying with rivals for ideas. It’s like with bookkeepers, he says. When PCs were introduced to the office, they began doing many of the tasks that employees used to do. But that didn’t lead to the disappearance of bookkeepers, rather they gradually turned into risk officers and portfolio managers. "They are now doing more challenging tasks. Once which make them and their employers more money."

That, Hoffman predicts, is how things will develop in many industries. Work may not become more varied and enjoyable for everyone, but it will for an increasing number of people. "And who knows? Maybe one day there will even be less of it."

Maybe not exactly the hammock. But not the welfare line, either.

^{Illustrationen: Pia Bublies / DER SPIEGEL}