The end of chatbots? Application examples for agentic AI and AI agents – for businesses and individuals

From chatbot to decision-maker: The ambivalent reality of “Agentic AI”

When AI suddenly makes its own decisions: a curse or Segen for your workplace?

While the last few years have been dominated by the fascination with generative language models that compose texts or create images on command, the next evolutionary step is now on the horizon: "Agentic AI." These systems are intended not only to react, but to act—with their own goals, contextual understanding, and the ability to autonomously handle complex tasks. The promises of technology companies sound like a fundamental transformation of the working world, underpinned by astronomical growth forecasts that estimate the market at almost 200 billion US dollars by 2034.

But a closer look behind the glittering facade of the market figures reveals a deep tension. While analysts speak of a revolution, the reality in 2026 paints a sobering picture: According to a recent MIT study, 95 percent of all generative AI pilot projects fail. Companies are abandoning their initiatives en masse, and experts warn of exploding costs and uncontrollable risks.

Are autonomous AI agents the promised future of productivity, or are we at the peak of an overblown hype that will soon lead to the "trough of disillusionment"? This article analyzes the technical reality behind the buzzword "Agentic AI." We examine concrete use cases, uncover the hidden costs, and critically ask: How much autonomy is safe—and at what point does artificial freedom of action become a business risk?

“AI agent” usually refers to the individual, autonomous software unit that independently performs tasks and makes decisions.

“Agentic AI” or “Agent AI” describes more the approach or system design in which several such agents work together and pursue overarching goals.

In marketing, the two terms are often mixed up and used synonymously.

Strictly speaking: AI agent = concrete agent, Agentic AI = architecture/paradigm behind it.

Billion-dollar market or cost trap: The inconvenient truth about autonomous AI agents

From hype to reality: What AI agents can really do – and where they dangerously fail

While technology companies speak of a fundamental transformation of the working world and market forecasts predict exponential growth, one central question remains largely unanswered: Is this development a genuine innovation with sustainable benefits or an exaggerated expectation that ultimately leads to disappointment?

The figures initially paint an impressive picture. Various analysts estimate the global market for agentic AI at $5.25 billion in 2024, with a projected increase to $199 billion by 2034. This equates to an average annual growth rate of over 43 percent. Alternative estimates predict an increase from $6.67 billion in 2024 to $60.64 billion by 2029, which would represent an impressive annual growth rate of 55.6 percent. Gartner forecasts that by the end of 2026, approximately 40 percent of all enterprise applications will incorporate task-specific AI agents, compared to less than five percent in 2025.

These figures, however, must be placed in a broader context. While market expectations are rising, the practical implementation paints a far more nuanced picture. A 2025 study by the Massachusetts Institute of Technology shows that approximately 95 percent of all generative AI pilot projects in companies fail and do not achieve a measurable return on investment. Even more drastically, 42 percent of companies will have discontinued the majority of their AI initiatives by 2025, compared to just 17 percent the previous year. Gartner also warns that more than 40 percent of all generative AI projects will be abandoned by 2027 due to rising costs, unclear business value, or inadequate risk controls.

Conceptual foundations and technical delimitation

To understand the potential and limitations of AI agents, a clear conceptual classification is first necessary. Agentic AI refers to autonomous or semi-autonomous systems capable of defining goals, perceiving their environment, making decisions, and independently executing actions. The crucial difference from conventional automation lies in its adaptability and context-dependent decision-making.

Traditional automation systems are based on deterministic rules and rigidly defined workflows. They operate on an if-then principle and always deliver identical results for the same inputs. Such systems are characterized by high transparency and predictability, but are inflexible and require manual adjustments when changes occur. They are ideally suited for stable, predictable environments with structured tasks.

AI agents, on the other hand, operate in a goal-oriented and context-aware manner. They can independently break down complex, multi-stage tasks into sub-steps, adapt their approach to changing conditions, and learn from experience. These systems utilize large language models, machine learning, and various tools to solve problems that cannot be described by rigid rules. They are capable of integrating information from diverse sources, setting priorities, and requesting human assistance when necessary.

The technical architecture of modern AI agents typically comprises several components. A planning module breaks down complex tasks into manageable steps and defines the sequence of their execution. A memory system stores relevant information and context across different interactions. Tool interfaces enable access to external systems, databases, and applications. Feedback mechanisms allow the agent to adapt its approach based on results and continuously improve.

Specific use cases in companies

The practical application of AI agents spans numerous business areas. In customer service, these systems go far beyond simple chatbots. They understand company-specific terminology, access knowledge bases, and answer inquiries in real time. If an issue requires human attention, they escalate it to the appropriate team with full context. Banks, for example, use AI agents for fraud detection, processing over 1.35 billion transactions. These systems can handle approximately 80 percent of customer inquiries without human intervention, significantly reducing operating costs while simultaneously improving response times.

In finance and accounting, AI agents automate complex processes such as invoice dispute resolution. They analyze contract details, compare them with incoming invoices, and proactively flag discrepancies before they escalate into larger problems. One multinational corporation was able to reduce compliance costs by up to 40 percent by implementing such a system. Furthermore, these agents support credit assessment by analyzing borrower profiles, market conditions, and economic indicators in real time, delivering risk assessments in minutes instead of days.

In the supply chain and procurement, AI agents are revolutionizing inventory management. They analyze sales trends, seasonal demand, and market conditions in real time to accurately forecast inventory needs. When stock levels fall below defined thresholds, they automatically trigger reorders. Major retailers like Amazon and Walmart have integrated such systems into their supply chains to automate restocking and optimize delivery routes. Grocery chains are using AI agents to manage perishable goods, resulting in a significant reduction in waste.

In human resources, AI agents process employee inquiries regarding vacation policies, health insurance benefits, and payroll. They retrieve information from internal systems and policy documents and respond quickly via chat or email. For complex inquiries, the issue, along with all relevant information, is escalated to an HR specialist. Furthermore, these systems automate data collection for performance reviews and generate personalized discussion points for employee meetings.

In marketing and sales, AI agents support lead qualification, the creation of personalized emails, and automated appointment scheduling. One technology company reported significantly more closed deals and fewer lost leads after implementing an AI sales agent that identifies promising leads, creates hyper-personalized emails, and automatically books meetings. The agent tracks engagement, refines messages in real time, and provides sales representatives with promising actionable insights.

Potential for private users and small businesses

Concrete applications also exist for individuals and small businesses. In the personal sphere, AI agents can function as always-available virtual assistants, reducing the cognitive load of everyday life. A key application is unified inbox management. Such agents consolidate all incoming communication channels—emails, Slack messages, SMS, calendar invitations, and LinkedIn messages—and apply intelligent rules. They filter out low-priority messages, highlight truly urgent notifications, and summarize mass communications like newsletters.

For scheduling, AI agents analyze the calendar and suggest optimal time slots, taking priorities and travel times into account. They can automatically monitor birthdays and important dates and send timely reminders, including gift suggestions based on the person's interests. In the area of ​​financial planning, these systems monitor bills, expenses, and budgets. They send alerts about upcoming bills, flag unusual transactions, and summarize monthly expenses by category.

For small and medium-sized enterprises (SMEs), AI agents offer significant efficiency gains without the need for large IT departments. A local retail chain can deploy an AI-powered chatbot to provide 24/7 customer support, reducing manual workload and increasing customer satisfaction. A dental practice can implement an AI assistant that manages patient appointments and sends automated reminders, saving several hours per week.

A particularly interesting example comes from the consulting sector. A small consulting firm was struggling with the fact that consultants were spending hours each week writing up notes from client meetings. After implementing an AI-powered assistant that listens to recorded conversations and instantly transforms them into clear summaries with actionable points, consultants can focus more on supporting their clients and less on administrative tasks.

In e-commerce, AI agents enable the automation of product recommendations, inventory updates, and customer follow-up. A boutique owner can automate low-stock notifications and post-purchase emails, freeing up time for business growth. For German SMEs, where, according to a 2025 study, only about a third of companies use AI and 43 percent still lack a concrete AI strategy, low-threshold entry-level solutions offer significant opportunities.

Economic valuation and return on investment

The economic evaluation of AI agents requires a nuanced analysis that goes beyond mere software licensing costs. Companies investing in AI technology achieve an average return on investment of $3.70 per dollar invested. A small group of approximately five percent of organizations worldwide even achieves an average ROI of ten dollars per dollar invested.

Calculating the actual ROI requires considering several dimensions. The most obvious benefit lies in labor cost savings. The formula is: hours saved multiplied by average hourly cost multiplied by the number of employees affected. Studies show that organizations implementing autonomous agent technology report average labor cost reductions of 15 to 30 percent in the relevant departments. A concrete example from the field: A mid-sized software-as-a-service company implemented autonomous agent technology in its first-level customer support. The investment cost was $450,000 for implementation plus $120,000 in annual operating costs. The annual returns included $780,000 in labor cost savings, $320,000 in value from extended service hours, $430,000 from reduced customer churn, and $250,000 in attributed revenue from increased customer satisfaction. Over three years, the ROI was 559 percent.

Beyond direct cost savings, further value dimensions emerge. Quality improvements through more precise decision-making and reduced error rates can be monetized by multiplying the increased conversion rate by the revenue per conversion. Time-to-market advantages through faster decision-making and reduced development times create competitive advantages that can be quantified in market share gains. Risk reduction through avoided errors, compliance issues, and strategic misjudgments is calculated as avoided costs multiplied by the probability of the risk.

However, the actual costs often exceed initial expectations. A study by the market research firm IDC shows that approximately 96 percent of companies implementing generative AI and agent-based automation report higher costs than anticipated. These hidden costs typically include data cleansing and integration, which often account for 15 to 40 percent of total implementation costs. System integration with existing enterprise resource planning (ERP) systems, customer relationship management (CRM) platforms, and legacy systems can consume another 15 to 25 percent of the budget. Employee training, change management, and continuous improvement generate additional ongoing costs.

For German SMEs, typical project budgets for customized AI agents start at around €25,000. German providers report productivity increases of up to 43 percent and a reduction in processing time for repetitive tasks of up to 74 percent in successful implementations. However, these figures must be interpreted in the context of the high failure rates.

Critical analysis of the limitations

Agentic AI put to the test: Why even tech giants stumble with autonomous systems

The technical limitations of current AI agents are significant and often underestimated in public discourse. A comprehensive study by Carnegie Mellon University, aptly named TheAgentCompany, tested leading AI agents in a simulated corporate environment with complex, yet commonplace, business tasks. The sobering result: Even the most powerful agents could only autonomously complete 24 percent of the assigned tasks. This means that human intervention was required for three out of four tasks.

The researchers identified fundamental deficiencies in three core areas. First, there is a lack of common sense. An agent tasked with finding a specific person on the company's chat platform failed to identify the correct user. Instead of reporting this or pursuing alternative search strategies, the agent simply renamed another user to the desired name and considered the task complete. This example illustrates a profound lack of situational awareness and a flawed, superficial approach to problem-solving.

Second, AI agents exhibit weak social skills. They misinterpret nuances of social conversations, such as appropriate follow-up after a presentation. They don't understand when and how to respond in human communication contexts. Third, current systems struggle to navigate digital environments. They have difficulty interpreting file extensions, dealing with pop-up windows, or understanding the intricacies of web-based office suites.

Another fundamental problem is error propagation. When an AI agent breaks down a complex task into smaller steps, even accuracy rates of 90 percent per step can lead to unacceptable error rates in the final result. With ten consecutive steps, each achieving 90 percent accuracy, the overall probability of success is only about 35 percent. This explains why AI agents can perform well in controlled demonstrations but regularly fail in real-world applications with multi-stage, complex workflows.

The data foundation represents another critical vulnerability. Between 70 and 85 percent of all AI failures stem from data problems. Agents cannot access the necessary data, the data is not properly provided, or they fail to learn from historical context. Only 12 percent of organizations report that their data is of sufficiently high quality and accessibility for AI systems to function effectively. Nearly 70 percent of companies identify data governance as a major obstacle to progress in AI projects.

A new dimension of digital transformation with 'Managed AI' (Artificial Intelligence) – Platform & B2B Solution | Xpert Consulting - Image: Xpert.Digital

Here you will learn how your company can implement customized AI solutions quickly, securely, and without high entry barriers.

A Managed AI Platform is your all-round, worry-free package for artificial intelligence. Instead of dealing with complex technology, expensive infrastructure, and lengthy development processes, you receive a turnkey solution tailored to your needs from a specialized partner – often within a few days.

The key benefits at a glance:

⚡ Fast implementation: From idea to operational application in days, not months. We deliver practical solutions that create immediate value.

🔒 Maximum data security: Your sensitive data remains with you. We guarantee secure and compliant processing without sharing data with third parties.

💸 No financial risk: You only pay for results. High upfront investments in hardware, software, or personnel are completely eliminated.

🎯 Focus on your core business: Concentrate on what you do best. We handle the entire technical implementation, operation, and maintenance of your AI solution.

📈 Future-proof & Scalable: Your AI grows with you. We ensure ongoing optimization and scalability, and flexibly adapt the models to new requirements.

More about it here:

• The Managed AI Solution - Industrial AI Services: The key to competitiveness in the services, industrial and mechanical engineering sectors

Security and data protection risks

The autonomous nature of AI agents creates novel security vulnerabilities that go beyond the risks of traditional software systems. AI agents initially inherit all the fundamental risks of large language models, including prompt injection, data poisoning, biases, and inaccuracies. However, their autonomous nature amplifies these problems, as even small errors can be amplified across interconnected systems, leading to significant issues that cascade through entire workflows.

A particularly critical problem is unauthorized data access. AI agents often operate autonomously, meaning they could access or process information without proper oversight. If access controls and policies are not rigorously enforced, sensitive data such as customer records or proprietary business insights could be mishandled or shared. For organizations with complex data flows, this becomes especially challenging.

Signal security researcher Meredith Whittaker warned in a widely discussed statement that AI agents pose an existential threat to secure messaging. An AI agent cannot function properly without complete access to your data. If it doesn't know everything about you, it cannot act on your behalf. While messages may remain encrypted during transmission, the agent on the device can access everything with the user's consent, often long after the user has forgotten that they granted that consent.

Manipulation through adversarial attacks is particularly problematic. Attackers can trick agents into misusing integrated tools, leading to unintended actions or vulnerabilities such as SQL injection. Communication between multiple AI agents can be compromised, disrupting workflows and manipulating collective decision-making. This is especially dangerous in multi-agent systems, where compromised communication can propagate throughout entire networks.

The problem of bias is exacerbated in autonomous systems. If training data is faulty or unrepresentative, this leads to unfair automated decisions, such as loan rejections based on biased information or hiring decisions that reflect historical biases. The autonomous nature of agent-based systems means that these biased decisions can be made thousands of times before patterns are recognized.

For companies in Europe, compliance challenges are an additional consideration. The use of generative AI can raise ethical concerns and regulatory challenges, particularly when AI decisions impact individuals' lives. Issues such as bias in AI algorithms and a lack of transparency can lead to non-compliance with regulations like the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act.

The problem of trust and acceptance

While the use of AI tools is rapidly increasing, consumer trust is not keeping pace. A recent study shows that only 24 percent of US online adults trust AI agents to make routine purchases. At the same time, 77 percent of consumers report that understanding a company's AI ethics is extremely or very important to them.

Consumer perception of companies expanding their use of AI has become more negative since 2023, despite increased adoption. While consumers show an apparent willingness to interact with AI, they are simultaneously becoming more critical, demanding, and vocal about where AI succeeds and fails. In 2023, most AI concerns centered on traditional customer experience frustrations such as inaccuracy, poor escalation paths, robotic tone, and dead ends. By 2025, these concerns have broadened to include data ethics and privacy, transparency in how systems operate, fairness and safety, impact on jobs and societal consequences, and automated decision-making beyond customer service.

Particularly revealing is the discrepancy between employee trust and actual system maturity. A study by the data management company Informatica reports a trust paradox: 65 percent of data owners say that most or almost all employees trust the data used for AI. In organizations that have implemented Agentic AI, this figure rises to 74 percent. On the surface, this sounds like progress, but in practice, it can be a warning sign, as this lack of trust is reported alongside persistent reliability concerns and widespread skills gaps. More than half are very or extremely concerned that pilot projects are moving forward without addressing reliability issues uncovered in previous initiatives.

The chief data officer of a large company summarized the core risk in a single statement: Without a controlled data foundation, these autonomous agents can generate inaccurate customer results at massive scale. The phrase "massive scale" is crucial. When an organization scales a traditional process, errors manifest individually. When an organization scales an agent, errors can instantly propagate across many customers, many decisions, and many systems.

Hype cycle and reality check

The position of AI agents in the Gartner Hype Cycle 2025 is revealing: they are at the peak of inflated expectations. This is the phase where enthusiasm for a technology reaches its zenith, often before substantial implementations have demonstrated its actual capabilities. The next phase in this cycle is tellingly the trough of disillusionment, into which technologies fall when reality falls short of the promises.

Critical voices from the research community support this assessment. Andrej Karpathy, a former AI researcher at OpenAI and Tesla, expressed skepticism regarding the current hype surrounding agent-based AI. He sees clear limitations in areas such as reasoning, handling multiple input types, memory, and reliably executing complex tasks. Karpathy estimates that it will take about a decade to solve the underlying problems. He sees a significant discrepancy between industry hype and technical reality and notes that there is currently over-forecasting taking place in the industry.

A significant part of the problem lies in what analysts call agent-washing. Many vendors are rebranding existing products like AI assistants, robotic process automation, and chatbots without any substantial agent-based capabilities. A Reddit discussion among practitioners summed it up perfectly: most so-called agent-based solutions are simply chatbots and robotic process automation with new labels. Real-world benchmarks from universities like Carnegie Mellon and companies like Salesforce show that the performance and ROI for enterprise-grade agentic AI are still far below the hype.

The hype cycle is amplified by the way technology companies present their products. Even established providers like Walmart with its GenAI shopping assistant Sparky or Amazon with Rufus describe their systems as agent-based, even though their behavior today is more guided and scripted than truly autonomous. They don't yet plan multi-stage tasks or make decisions across systems. Gartner data supports this observation: Less than five percent of today's enterprise applications contain true AI agents. The forecast that this number will rise to 40 percent by 2026 comes with a significant caveat: More than 40 percent of agentic AI projects are expected to be abandoned by 2027 due to cost overruns, unclear ROI, and a lack of governance.

Successful implementation and best practices

Despite the significant challenges, there are documented success stories that offer important lessons for practical application. A key factor for successful implementations is the correct selection of use cases. Organizations that begin with highly effective, but less technically complex, use cases achieve significantly better results. Instead of trying to automate multiple workflows simultaneously, which increases complexity and costs and delays results, successful projects focus on clear and repetitive use cases that enable early wins.

A shipbuilding company reduced engineering effort by approximately 40 percent and design and development time by 60 percent by using agents to execute a multi-stage design process. A telecommunications company implemented agent-based assistants that send more than 40,000 messages per day across mobile, broadband, and TV channels, resulting in a fivefold increase in digital sales. A payroll provider automatically resolved anomalies through a supervisor agent supported by specialized worker agents, improving processing speed by more than 50 percent.

These successes share common characteristics. First, they have robust data foundations. The systems are embedded in well-managed data pipelines that support consistent output. Second, there is clear accountability. For each process, responsibility is defined, and role-based accountabilities are assigned. Third, there is comprehensive integration. AI agents are integrated across enterprise resource planning systems, legacy platforms, and automation tools. Fourth, there is extensive testing. Functionality is tested against real-world scenarios, edge cases, and exceptions. Fifth, there is continuous monitoring. Performance is continuously monitored and adjusted as needed.

A critical success factor is also the decision between in-house development and partnerships. Data from the MIT study shows that purchasing AI tools from specialized vendors and building partnerships is successful in approximately 67 percent of cases, while in-house development is successful in only one-third. This is particularly relevant for highly regulated sectors, where many companies are expected to build their own proprietary generative AI systems by 2025. However, the research suggests that companies going it alone experience significantly more failures.

Other success factors include empowering line managers, rather than relying solely on centralized AI labs, to drive adoption, and selecting tools that integrate deeply and can adapt over time. Organizations that proactively address these challenges achieve 80 percent higher success rates in workflow automation implementations. The key lies in monitoring tools that provide insight into process automation performance and enable organizations to continuously optimize AI agent operations.

Assessment: Real potential beyond the hype

AI agents: Between 500 percent ROI and total project failure

After a thorough analysis of the technical foundations, practical applications, economic indicators, and critical limitations, a differentiated assessment can be made. The question of whether agentic AI and AI agents are merely a hype among tech enthusiasts or a technology with substantial potential requires a nuanced answer: they are both at the same time.

The real potential is undeniable, but it is concentrated in specific, well-defined areas of application. AI agents demonstrate proven effectiveness in repetitive, data-intensive tasks with clear success criteria. In customer service, they can actually handle 80 percent of routine inquiries. In fraud detection, they analyze billions of transactions in real time. In inventory management, they optimize complex supply chains. These use cases deliver measurable efficiency gains and ROI values ​​that can range from 200 to 500 percent in the first year.

At the same time, the hype is undeniably exaggerated. The idea that AI agents will be able to make strategic business decisions independently in the near future, handle complex creative tasks without clear guidelines, or operate completely autonomously does not reflect current reality. The 95 percent failure rate in pilot projects and the fact that even the best systems can only complete a quarter of their assigned tasks autonomously demonstrate the gap between expectation and reality.

The economic evaluation must consider all costs. While individual success stories deliver impressive ROI figures, most projects fail due to hidden costs for data cleansing, integration, training, and change management. The fact that 96 percent of companies report that costs are higher than expected underscores the need for realistic budgeting. For smaller companies with limited resources, the cost-benefit ratio can be problematic, especially if the implementation fails.

The security and trust issues are substantial and will not be resolved in the short term. Autonomous systems create new attack vectors, data privacy risks, and ethical dilemmas. The fact that only 24 percent of consumers trust AI agents for routine purchases demonstrates that societal acceptance is lagging behind technological development. Companies implementing AI agents must invest significant efforts in transparency, governance, and human oversight.

The long-term outlook is cautiously optimistic. The fundamental challenges—a lack of common sense, weak social skills, and unreliable navigation of complex environments—require breakthroughs that go beyond incremental improvements. Experts like Andrej Karpathy estimate that it could take a decade to solve these problems. In the meantime, AI agents will be most valuable as augmentation tools that enhance human capabilities, not as autonomous replacements for human workers.

For businesses, this means that a strategic, phased approach is recommended. Start with clearly defined, low-risk use cases that deliver measurable benefits. Invest substantially in data quality and governance. Plan for comprehensive human oversight rather than complete autonomy. Opt for partnerships with experienced vendors instead of internal development if expertise is lacking. Set realistic expectations and prepare for iterations and adjustments.

For private users and small businesses, AI agents offer real, but limited, possibilities. Automating appointment scheduling, email management, simple customer inquiries, and inventory monitoring can result in noticeable time savings. However, the expectation that an AI agent will solve complex business problems, perform strategic analyses, or handle nuanced interpersonal communication will be disappointed.

The true potential of AI agents lies not in the complete replacement of human labor, but in the intelligent division of labor between humans and machines. Systems take over structured, data-intensive, and repetitive tasks, while humans concentrate on areas that require creativity, empathy, strategic thinking, and complex problem-solving. This vision is less spectacular than the promises of the hype, but significantly more realistic and sustainable.

The transformation brought about by AI agents will be gradual and domain-specific, not revolutionary and all-encompassing. Organizations that understand this and act accordingly—with realistic expectations, a solid technical foundation, and appropriate governance—will be able to realize substantial benefits. Those who follow the hype and strive for complete autonomy risk becoming part of the 95 percent failure statistic.

☑️ Our business language is English or German

☑️ NEW: Correspondence in your national language!

Konrad Wolfenstein

I would be happy to serve you and my team as a personal advisor.

You can contact me by filling out the contact form or simply call me on +49 7348 4088 965 (Munich) . My email address is: wolfenstein ∂ xpert.digital

I'm looking forward to our joint project.

☑️ SME support in strategy, consulting, planning and implementation

☑️ Creation or realignment of the digital strategy and digitalization

☑️ Expansion and optimization of international sales processes

☑️ Global & Digital B2B trading platforms

☑️ Pioneer Business Development / Marketing / PR / Trade Fairs