Smoke & Mirrors

Servinomy is hunting the facts behind the claims within service management. No judgment, just sharp investigation and recording of the current culture in the business. For each subject, we publish directional guidance rather than a solution.

Agentic AI

Silver Bullet or Wishful Thinking?

By Servinomy | July 2026

Introduction: The Promise of Agentic AI

It is 2026. In every boardroom, at every technology conference, and across every LinkedIn feed, the same mantra resonates: "Agentic AI will change everything." Autonomous AI agents that independently make decisions, execute complex workflows, and resolve IT incidents without human intervention — the promises are grand, the expectations sky-high.

Yet anyone familiar with the history of technology recognises this pattern immediately. We have seen it before: with ERP systems in the 1990s, with Big Data in the 2010s, with blockchain, with the Internet of Things, with the metaverse. Time and again, a new technology is presented as the silver bullet — the wonder cure that will solve all problems. And time and again, disillusionment follows.

The question this article poses is not whether Agentic AI has any value — it undoubtedly does in specific contexts. The question is: is it being sold as the latest technological silver bullet, whilst the reality is fundamentally more complex? And what does this mean specifically for the ITSM world?

The very definition of Agentic AI is telling. Gartner describes AI agents as "autonomous or semi-autonomous software entities that use AI techniques to achieve goals in their digital or physical environment"[4]. It sounds impressive. But as we shall see, behind this definition lies a world of nuance, limitations, and unresolved challenges.

The Anatomy of a Technology Hype

Amara's Law and the Gartner Hype Cycle

Roy Amara, former president of the Institute for the Future, formulated a law in the 1960s that has been proven correct time and again: "We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run."[5] This law — known as Amara's Law — is the key to understanding the Agentic AI hype.

In August 2025, Gartner explicitly placed AI agents at the "Peak of Inflated Expectations" of the Hype Cycle for Artificial Intelligence [4]. This is the point at which expectations are furthest removed from reality — the moment just before the inevitable "Trough of Disillusionment".

Eric Siegel, former Columbia University professor and author of The AI Playbook, goes further in his critique. Writing in Forbes, he argues that "agentic AI" does not refer to any specific technological methodology or breakthrough, but is merely a rebrand of existing generative AI ambitions. He calls it outright "the new vaporware"[6] — software that is promised but never delivered.

Historical Parallels: A Pattern of Repetition

The history of IT is littered with technologies presented as silver bullets. A comparison is revealing:

ERP (1990s)
Big Data (2010s)
Blockchain (2017-2019)
Metaverse (2021-2022)
Agentic AI (2024-present)

Technology

The Promise
The Reality

ERP (1990s)

(P) Integrated business automation, end of silos
(R) Massive implementation failures, cost overruns, new silos

Big Data (2010s)

(P) Collect everything and insights will emerge
(R) Unused dashboards, data quality problems, high costs

Blockchain (2017-2019)

(P) Trustless systems will replace institutions
(R) Speculative collapse, niche applications, heavy regulation

Metaverse (2021-2022)

(P) We will live and work in virtual worlds
(R) Massive write-offs, limited use, training & simulation only

Agentic AI (2024-present)

(P) Autonomous agents will fully replace human work
(R) 40%+ projects cancelled, only 130 genuine agents, return to humans

Fred Brooks, Turing Award winner and author of The Mythical Man-Month, wrote his famous essay "No Silver Bullet — Essence and Accident in Software Engineering" as far back as 1986. His central thesis: there is no single technology or management technique that can promise a tenfold improvement in productivity, reliability, or simplicity. Forty years later, this warning is more relevant than ever.

"Agent Washing": The New Greenwashing

A particularly concerning phenomenon is what Gartner calls "agent washing": companies simply rebranding existing technologies — chatbots, RPA (Robotic Process Automation), virtual assistants — as "Agentic AI" without making substantive changes [2]. Of the thousands of so-called "AI agent" providers assessed by Gartner, only approximately 130 exhibited genuine agentic capabilities. This represents less than 1% of the total market offering.

Anushree Verma, Senior Director Analyst at Gartner, states plainly: "Most agentic AI propositions lack significant value or return on investment, as current models don't have the maturity and agency to autonomously achieve complex business goals or follow nuanced instructions over time."[2]

Agentic AI in Practice: Facts vs Fiction

The Hard Numbers

The data is sobering. A Gartner survey of 3,412 participants in January 2025 revealed [2]:

Only 19% reported significant investment in AI agents
42% were investing conservatively
31% adopted a wait-and-see approach
8% had not invested whatsoever

A separate Gartner study of 360 IT leaders (October 2025) revealed the gap between hype and reality even more sharply [7]:

Only 15% are considering, piloting, or deploying fully autonomous agents
Only 19% have high or complete trust in their vendor's ability to protect against AI hallucinations
74% view AI agents as a new attack vector within their organisation
Only 7% believe AI agents will replace human workers within the next four years

Case Study: Klarna and Duolingo — The Return of the Human

No example better illustrates the gap between promise and reality than Klarna. The Swedish fintech giant announced with great fanfare in 2023 that AI had taken over the work of 700 customer service agents. CEO Sebastian Siemiatkowski declared in December 2024: "AI can already do all the jobs that we, as humans, do."

Two years later, Siemiatkowski tells a different story. Klarna is preparing to hire more human workers after the quality of AI customer service failed to meet expectations. The CEO openly acknowledged: "Cost unfortunately seems to have been a too predominant evaluation factor when organising this. What you end up having is lower quality."[3]

Duolingo followed a comparable path. The language-learning application began replacing contractors with AI, but encountered similar quality problems. Starbucks and McDonald's — which implemented AI systems for customer interactions — likewise returned to human workers [3]. This pattern of "AI-layoff reversal" has become a recognised industry phenomenon.

The Hallucination Problem: A Fundamental Limitation

One of the most fundamental limitations of Agentic AI is the hallucination problem. Research demonstrates that AI outputs can contain up to 40% factual errors in some domains [8]. A comparative study by Chelli et al. (2024) showed hallucination rates of 39.6% for GPT-3.5 and 28.6% for GPT-4 — and these are the most advanced models currently available.

For an autonomous AI agent independently making decisions in a business environment, these percentages are unacceptably high. Imagine: an AI agent autonomously resolving IT incidents, but arriving at an incorrect diagnosis in nearly 30% of cases. Or an agent managing customer communications, but regularly providing inaccurate information. The consequences for businesses could be catastrophic.

This explains why 71% of IT professionals still manually verify AI outputs, and 62% report difficulty trusting AI recommendations [9]. In an environment where speed and reliability are critical — such as ITSM — this is a fundamental problem.

The Five Most Common Failure Patterns

Research by Tactical Edge (2026) identifies five structural failure patterns in Agentic AI implementations:

1. Autonomy without clear boundaries: Agents granted excessive freedom without defined constraints lead to unpredictable behaviour that conflicts with business rules and regulations.

2. Missing observability: Without comprehensive logging and monitoring, debugging agentic systems is virtually impossible.

3. Over-reliance on prompts: Systems that work in testing fail in production due to context changes and edge cases.

4. Governance as an afterthought: Retrofitting compliance requirements is significantly more difficult than integrating them from the outset.

5. Demo-driven development: Projects optimised for impressive demonstrations rather than production reliability.

The ITSM Dimension: Promise and Reality

Where AI in ITSM Does Work

It would be unfair to highlight only the failures. Research by TeamDynamix (2026) demonstrates that organisations correctly implementing AI in ITSM achieve measurable results [10]:

82% report ticket deflection (fewer tickets requiring human handling)
71% experience faster resolution times
76% report improved customer satisfaction
61% say AI has accelerated root cause analysis

Notably, the most successful use of AI in ITSM is not the autonomous agent, but knowledge management and content generation — cited by 88% of successful implementations. This is precisely the opposite of what the hype promises: not autonomous decision-making, but support for human knowledge workers.

The Complexity Paradox in ITSM

SolarWinds' IT Trends Report 2026 reveals a paradox characteristic of the current state of AI in ITSM [9]: AI was supposed to make ITSM simpler. In reality:

44% of IT professionals say that managing incident response has become a new or increased responsibility because of AI adoption
Only 27% report any meaningful reduction in alert volume thanks to AI
41% of first-line managers say AI has raised expectations without reducing workload

Doug Murray, CEO of Auvik, summarises it aptly: "When three-quarters of IT leaders believe they have an AI policy, but fewer than half of help desk staff say the same, that's an implementation problem, not a policy problem."[10]

McKinsey confirms this picture: flooding an organisation with AI agents rarely works. Success depends on fundamentally rethinking workflows and correctly integrating the technology [11]. Only 32% of leaders report sustained, enterprise-wide AI impact.

The Architecture Gap: Native vs Bolt-on AI

A crucial distinction the ITSM industry is learning is the difference between AI-native and bolt-on AI. Research by OpenText demonstrates that organisations using public LLMs without platform integration achieve measurably worse results than those running AI natively within their ITSM environment. Vendors such as ServiceNow, Ivanti, and Workday have all reached the same conclusion: AI built into the platform outperforms AI added on top of it.

Security, Governance and Ethics

OWASP and the Security Risks of Autonomous Agents

OWASP published the State of Agentic AI Security and Governance 2.0 report in December 2025 [12], identifying two critical risks as the highest priority for enterprise deployments:

Tool poisoning: Malicious or compromised tools that manipulate agent behaviour
Multi-agent coordination failures: Trust breakdowns between agents leading to unintended or harmful outcomes

OWASP advises organisations to abandon reliance on model-based benchmarks and instead adopt system behaviour evaluation — assessing what the full deployed agent stack actually does at runtime. This represents a fundamentally different approach from current market practice.

Regulatory Challenges: The EU AI Act and Agentic AI

The EU AI Act imposes requirements on high-risk AI systems about behaviour and auditability. Agentic AI implementations may not satisfy these requirements under current model-only evaluation approaches, creating direct compliance risks for European organisations. This is a dimension that is almost absent from the hype discourse.

Conclusion: Beyond the Silver Bullet

Is Agentic AI the latest technological silver bullet? The answer is nuanced — but the nuance itself is telling.

Yes, Agentic AI in its current form is being sold with promises that far exceed technological reality. The numbers are clear: 40%+ of projects cancelled, only 130 genuine agents from thousands of providers, companies returning to human workers, and a fundamental hallucination problem that makes autonomous decision-making unreliable in critical environments.

But Amara's Law reminds us that we overestimate technology in the short run and underestimate it in the long run. The genuine value of AI in ITSM — knowledge management, ticket deflection, root cause analysis — is real and measurable, provided implementation is correct.

The lesson for ITSM professionals is clear: be sceptical of the hype, but not of the technology itself. Demand concrete ROI evidence from vendors, not impressive demos. Invest in the fundamentals — data quality, governance, integration — before deploying autonomous agents. And remember: technology does not solve organisational problems. People do.

"The next decade will not belong to the companies with the smartest models. It will belong to the companies that can say: 'Yes, this system still works on a bad day — and here is why.'"

— Simplico, After the AI Hype (2026)

References

[1] Rashidi, S. (2025, 28 June). AI Agents And Hype: 40% Of AI Agent Projects Will Be Cancelled By 2027. Forbes. https://www.forbes.com/sites/solrashidi/2025/06/28/ai-agents-and-hype-40-of-ai-agent-projects-will-be-canceled-by-2027/

[2] Maruccia, A. (2025, 30 June). Agentic AI is all hype for now, says Gartner. TechSpot. https://www.techspot.com/news/108499-gartner-warns-agentic-ai-projects-fail.html

[3] Economic Times. (2025, 19 May). Company replaces 700 employees with AI; two years later, it's rehiring humans as AI falls short. The Economic Times. https://economictimes.indiatimes.com/news/new-updates/company-replaces-700-employees-with-ai-two-years-later-its-rehiring-humans-as-ai-falls-short/articleshow/121263692.cms

[4] Gartner. (2025, 5 August). Gartner Hype Cycle Identifies Top AI Innovations in 2025. Gartner Newsroom. https://www.gartner.com/en/newsroom/press-releases/2025-08-05-gartner-hype-cycle-identifies-top-ai-innovations-in-2025

[5] Chugani, V. (2026, 14 April). Amara's Law Explained: Why Our Technology Forecasts Are Systematically Wrong. Statology. https://www.statology.org/amaras-law-explained-why-our-technology-forecasts-are-systematically-wrong/

[6] Siegel, E. (2025, 28 July). The Agentic AI Hype Cycle Is Out Of Control — Yet Widely Normalised. Forbes. https://www.forbes.com/sites/ericsiegel/2025/07/28/the-agentic-ai-hype-cycle-is-insane--dont-normalize-it/

[7] Speed, R. (2025, 1 October). AI agent hypefest crashing against cautious leaders: Gartner. The Register. https://www.theregister.com/software/2025/10/01/ai-agent-hypefest-crashing-against-cautious-leaders-gartner/1535992

[8] Gomez, F. (2025, 11 november / 11 November). The 2025 AI Reality Check: How Businesses Are Winning the War on Hallucinations. Notrus AI. https://notrus.ai/insights/ai-hallucinations

[9] Nolan, S. (2026, 22 April). More AI, More Complexity: What SolarWinds' 2026 Report Really Means for IT Service Management. UC Today. https://www.uctoday.com/service-management-connectivity/ai-incident-management-it-service-management-trends-2026/

[10] Law, M. (2026, 29 May). AI in ITSM Is Delivering Real Results. So Why Are Most IT Teams Still Stuck? UC Today. https://www.uctoday.com/service-management-connectivity/ai-in-itsm-is-delivering-real-results-so-why-are-most-it-teams-still-stuck/

[11] Simplico. (2026, 5 January). After the AI Hype: What Always Comes Next (And Why It Matters for Business). Simplico. https://simplico.net/2026/01/05/after-the-ai-hype-what-always-comes-next-and-why-it-matters-for-business/

[12] AI Governance Institute. (2026, 17 June). OWASP Agentic AI Report Flags Tool Poisoning and Multi-Agent Failures as Top Enterprise Governance Gaps. AI Governance Institute. https://aigovernance.com/news/owasp-agentic-ai-report-flags-tool-poisoning-and-multi-agent-failures-as-top-enterprise-governance-gaps

A Sector at a Crossroads

The dominant online narrative.

By Servinomy | June 2026

Anyone searching for "service management" online today is immediately confronted with an overwhelming technological narrative. Automation, artificial intelligence, cloud platforms and digital transformation set the agenda.

Their collective message revolves around efficiency, scalability, compliance and automation. Optimising processes, reducing tickets, standardising workflows and improving KPIs: this is the language of the sector.

Service management stands at a crossroads. The dominant online narrative — technology, automation, efficiency — is not wrong, but it is incomplete. It answers the question of how we can optimise service management, but sidesteps the more fundamental question: why?

The emerging counter-movement — human-centred, purpose-driven, focused on value creation — offers a necessary counterweight. But as long as this movement remains underrepresented in the online discourse, organisations will continue to invest in systems that optimise processes without asking the underlying question: for whom, and to what end?

The challenge for service management professionals is therefore not primarily technological. It is a narrative challenge: the ability to tell a compelling story about the human and societal value of good service delivery — a story powerful enough to complement and correct the platform discourse.

We've Known the Answer Since 1964.

So Why Are We Still Failing?

By Servinomy | June 2026 - DEMO version

In 1964, Harold Leavitt told us that changing technology without changing people and process would make the system fight back. The service management community turned this into the People–Process–Technology (PPT) framework. We built ITIL around it. We certified thousands of practitioners on it.

In 2025, Gartner surveyed 3,100 CIOs across 88 countries and found that only 48% of digital transformation initiatives meet their targets. VML's global report found 64% of projects launch without a clear roadmap. Bain & Company put the broader failure rate for transformation at 88%.

Key Takeaways

The PPT model is not a relic — it is a diagnosis the industry keeps refusing to act on.
Organisations consistently over-invest in Technology, underfund Process, and treat People as a training line item.
Every generation of technology (ERP, Cloud, AI) repackages the same failure with new vocabulary.
The organisations that succeed treat people and business leaders as equally accountable — not just IT.

Is ITIL Dead?

The Answer Is More Complicated Than You Think.