AI in Business: What the Hype Conceals – and What Mid-Sized Companies Must Do Now
10 min Read Time
The Key Takeaways
- Klarna as a warning: Klarna fired 700 customer service employees due to AI – and later rehired them after service quality collapsed. The case shows: AI implementation without process understanding fails.
- AI agents don’t replace processes: AI agents (OpenAI Operator, Google Project Mariner, Microsoft Copilot Studio) will be production-ready in 2026 and intervene far more deeply in business processes than any chatbot.
- Mid-sized companies need a different strategy: Token costs for agentic systems are 5 to 10× higher than for simple chatbots – chronically underestimated in budgets.
- Digital foundations first: Whoever doesn’t understand and document their processes cannot deploy AI in a controlled way.
- The gap widens in 2026: The divide between AI pioneers and laggards will widen faster than ever before – whoever delays action now will fall behind irreversibly.
700 employees laid off, then rehired – and market capitalization still dropped by billions. What Klarna experienced with its AI experiment is no isolated incident but a warning signal for every company treating AI as a quick fix for cost pressure. While U.S. tech circles debate whether AI will take over all cognitive work within one to five years, German mid-sized companies struggle with cloud migration, process documentation, and IT skills shortages. These two worlds appear distant. They aren’t.
Because AI development is reshaping competitive conditions for everyone engaged in knowledge work – regardless of whether a company employs 50 or 50,000 people. Token costs for agentic AI vs. simple chatbots (OpenAI, 2025) underestimate this dynamic. Adopting it uncritically leads straight into the Klarna trap.
Two Worlds, One Issue
The debate between investor Matt Shumer and neuroscientist Gary Marcus is, at its core, an American one. In his widely discussed essay “Something Big Is Happening” – picked up by Fortune, CNN, and CNBC – Shumer claimed AI would fully assume cognitive work within one to five years. Marcus countered with the charge of “weaponized hype.” Two extreme positions – both missing the lived reality of European mid-sized businesses.
The German mid-sized sector operates in a different reality. Many companies are still asking fundamental questions: How do we migrate on-premise systems to the cloud? How do we digitize processes that have run on paper and Excel for twenty years? How do we find IT specialists capable of delivering this transformation? The idea that AI will replace entire departments within a few years feels almost surreal in this context.
Yet therein lies the danger. The speed at which AI capabilities improve affects not only Silicon Valley startups – it reshapes competitive conditions for every company performing knowledge work – and that includes nearly all mid-sized firms. Ignoring this risks far more than a short-term efficiency lag. Which structural patterns will define industry in 2026 is analyzed in the article Decentralized Intelligence as a Leitmotif: What Will Shape Industry in 2026.
What Klarna Teaches Us About the Risks of Hype
To understand what happens when companies translate AI hype uncritically into operational decisions, look to Klarna. For months, the Swedish payment service provider was hailed as a textbook example of successful AI integration. CEO Sebastian Siemiatkowski publicly announced that AI chatbots had taken over the work of 700 customer service staff. The workforce was drastically reduced; new hires were halted.
Disillusionment followed swiftly. Service quality dropped measurably. Customers complained about erroneous responses, unresolved issues, and the impossibility of reaching a human agent. Klarna backtracked – and began rehiring human staff. In a labor market it had itself thinned out.
The Klarna case is instructive because it reveals a pattern set to repeat across industries. Companies positioning AI as a direct replacement for human labor, rather than as a tool to augment it, walk straight into a trap. The technology impresses in demos – but remains insufficiently reliable to handle complex, context-dependent tasks sustainably without human oversight. That holds true as much today, early in 2026, as it did at the time of Klarna’s misstep.
For decision-makers in mid-sized companies, there’s a vital lesson here: The question isn’t whether to deploy AI. It’s how – and with what expectations. Why many companies hesitate despite such warning signs is explained in the article AI in the Mid-Sized Sector: Why So Many Companies Hesitate – and What Matters Now.
Current Development in 2026: AI Agents Shift the Debate
By early 2026, the landscape has shifted fundamentally once again. Where large language models and chatbots previously dominated the discussion, autonomous AI agents now stand center stage. Systems like Devin (Cognition AI) or Microsoft Copilot Studio enable AI not just to respond – but to execute multi-step tasks independently: writing code, running tests, generating documentation.
Simultaneously, OpenAI’s Operator feature powers autonomous browser agents that, on users’ behalf, fill out forms, make bookings, and conduct web research. Google DeepMind followed suit with Project Mariner. What was considered experimental territory in 2024 is now production-ready in 2026 – and reaching mid-sized enterprise environments.
This significantly alters the risk calculus. Agentic AI systems intervene far deeper in corporate processes than a simple chatbot. Errors escalate faster, since no human reviews each intermediate step. At the same time, efficiency gains multiply for companies deploying these systems thoughtfully. The gap between early adopters and wait-and-see players widens even faster than the Shumer-Marcus debate anticipated.
For the mid-sized sector, the core thesis – lay foundations first, then automate – holds truer than ever in 2026. Whoever fails to understand and document their processes cannot deploy agentic AI in a controlled manner. Uncontrolled deployment ultimately costs more than the original digitalization deficit.
Cloud Budgets Under New Pressure
The AI debate directly impacts IT budgets and cloud strategies. Major hyperscalers – Microsoft, Google, Amazon – are increasingly bundling their AI services with their cloud platforms. To use GPT-4o or comparable models productively, organizations can hardly avoid Azure, Google Cloud, or AWS. This dramatically changes the calculation for cloud migrations.
Many companies planned their cloud budgets based on assumptions one or two years old. Since then, AI services have emerged as a cost factor entirely absent from those original calculations. API calls to large language models, fine-tuning proprietary models on corporate data, infrastructure needed for data security and compliance – all drive costs upward. With agentic systems, token consumption per task rises sharply: industry observers estimate the added expense at 5-10× that of simple chatbot interactions.
At the same time, pressure mounts on IT departments to deliver AI projects rapidly. CEOs who’ve read Shumer’s essay – or similar pieces – expect results. The gap between expectation and feasibility widens – and IT leaders find themselves in the unenviable position of having to deliver innovation while holding the line on budgets.
A realistic cloud strategy must therefore explicitly price in AI costs – not as a vague line item, but as a defined budget category with specific use cases, measurable goals, and clear exit criteria. Companies launching AI projects without this structure risk budget surprises that jeopardize other strategic investments.
Security Becomes More Complex, Not Simpler
One aspect chronically underexposed in public AI debates: the security dimension. Every AI service integrated into corporate processes expands the attack surface. Large language models can be manipulated via so-called prompt injection attacks. Corporate data used for training or operating AI models must be protected – not only from external attackers but also from the providers themselves.
Regulatory requirements are tightening in parallel. The EU AI Act is entering into force incrementally and obliges companies to classify, document, and – in certain cases – audit their AI usage. For the mid-sized sector, this means additional effort in an area already thinly staffed.
Security professionals face the challenge of integrating AI systems into existing security architectures without letting complexity spiral uncontrollably. That demands not only technical expertise but also the ability to communicate clear boundaries to executive leadership. Not every AI use case is security-justifiable – and the capacity to say “no” with sound justification is becoming an underappreciated leadership competency. Even Lloyd’s of London now offers specific insurance against losses caused by AI hallucinations – a sign that the risk industry takes this dimension extremely seriously. More on this: Lloyd’s of London Offers Insurance Against Losses from AI Hallucinations.
What Leadership Means Now
The real question behind the Shumer-Marcus debate isn’t technological. It’s strategic: How do leaders make decisions amid extreme uncertainty? Neither optimists nor skeptics can reliably predict how capable AI systems will be in three or five years. Development proceeds neither linearly nor predictably.
For CEOs and executive boards, this means cultivating a stance that avoids both blind activism and passive waiting. Concretely: launch pilot projects – but with clear success criteria. Invest in building internal competence before outsourcing judgment to external consultants. Upskill employees instead of prematurely substituting them with technology. And: understand your own value creation before automating it.
Companies unable to clearly describe their core processes won’t optimize them meaningfully with AI either. Digitizing the foundations – structured data, documented processes, modern IT infrastructure – remains the prerequisite for everything that follows. Skipping this step and jumping straight to AI builds on sand. What this implies for the strategic agenda in 2026 is discussed in the article Business Trends 2026: How Companies Shape the Next Phase of AI.
“Most agentic AI projects today are early-stage experiments, frequently deployed incorrectly. That can blind enterprises to the real costs and complexity involved in scaling.”
– Anushree Verma, Senior Director Analyst, Gartner, June 2025
The Gap Widens
The most uncomfortable truth of today’s debate: The divide between companies strategically deploying AI and those still wrestling with basic digital infrastructure will dramatically widen in the coming years. This affects not only efficiency and costs – but also the ability to attract talent, retain customers, and remain competitive in global markets.
Matt Shumer may be overly optimistic in his timelines. Gary Marcus may be right that current technology faces fundamental limits. But the direction of development is unmistakable – and accelerating. For the German mid-sized sector, this means: act now – but act wisely. Build foundations, understand risks, bring your people along. Don’t let hype drive you – or skepticism paralyze you.
That’s no easy task. But it’s precisely the kind of challenge that defines sound corporate leadership.
- Klarna laid off roughly 700 customer service employees due to AI – and rehired them after service quality collapsed.
- AI agents are the megatrend of 2026: OpenAI Operator, Google Project Mariner, and Microsoft Copilot Studio autonomously execute multi-step tasks.
- Token costs for agentic AI are 5-10× higher than for simple chatbot interactions – an often underestimated budget factor.
- EU AI Act: Phased implementation since 2024; full compliance obligations for high-risk applications begin in 2026.
- Lloyd’s of London already offers insurance against losses caused by AI hallucinations – the risk industry has recognized the threat.
Frequently Asked Questions
What triggered the AI debate between Shumer and Marcus?
Matt Shumer’s essay “Something Big Is Happening” predicted AI would assume cognitive work within one to five years. Gary Marcus criticized these claims as exaggerated hype lacking sufficient empirical evidence. The debate was picked up by Fortune, CNN, and CNBC – and has since polarized the expert community.
Why does the AI debate matter to the German mid-sized sector?
Even though many mid-sized firms are still focused on foundational digital work, AI is rapidly transforming competitive conditions. Companies ignoring the topic risk falling further behind in efficiency, talent acquisition, and customer retention – even if the most extreme forecasts never materialize.
What does the Klarna case teach us about AI in customer service?
Klarna replaced roughly 700 customer service staff with AI chatbots – and later rehired human personnel after service quality plummeted. The case demonstrates that deploying AI as a replacement – not as an augmentation – of human work carries significant risks – and can prove more expensive long-term than maintaining the status quo.
What are AI agents – and why are they so relevant in 2026?
AI agents – such as OpenAI Operator, Microsoft Copilot Studio, or Google Project Mariner – execute multi-step tasks autonomously, rather than answering single questions. They intervene more deeply in corporate processes, escalate errors faster, and incur substantially higher token costs than classic chatbots.
How do AI agents affect budget planning from 2026 onward?
Agentic AI systems generate significantly higher token costs than simple chatbots – industry observers estimate 5-10× the expense. Companies failing to account for these additional costs risk unpleasant budget surprises that threaten other strategic investments.
What security risks accompany AI deployment?
Every integrated AI service expands a company’s attack surface. Prompt injection attacks can manipulate language models, and corporate data must be protected both from external attackers and from AI providers themselves. The EU AI Act introduces additional documentation and audit obligations – creating extra workload in mid-sized firms already short on personnel.
What should mid-sized companies do first to prepare for AI?
The most critical prerequisite is digitizing the foundations: structured data, documented processes, and modern IT infrastructure. On that basis, targeted pilot projects – with clear success criteria and defined budgets – can be launched. Agentic systems come only afterward.
Further Reading
- Agentic AI in the Mid-Sized Sector – MyBusinessFuture
- Private Cloud for AI – cloudmagazin
- Change Management in AI Transformation – Digital Chiefs
Header Image Source: Pexels / Pixabay
