The promise was immense: that of autonomous, proactive artificial intelligence, capable of acting in a contextualized manner and profoundly transforming organizations. Two years later, the enthusiasm confronts a harsher reality. On June 25, 2025, Gartner released a straightforward prediction. More than 40% of Agentic AI projects will be abandoned by the end of 2027, due to high implementation costs and difficult-to-measure business value. A strong signal for companies, too often tempted by publicity stunts rather than structural transformation.

Agentic AI, Between Hype and Disappointment

Agentic AI—these systems capable of making decisions, taking action, and learning with increasing degrees of autonomy—has established itself as the new frontier of generative AI. Since late 2023, laboratories and startups have been competing with prototypes and demonstrators: multitasking conversational agents, legal assistants, business copilots, and even software development agents.

But according to Gartner, this excitement masks a profound imbalance. The majority of projects launched in the past 18 months have not moved beyond the experimental stage, and very few manage to demonstrate repeatable performance in production. Worse still: only 130 providers worldwide currently offer truly agentic solutions according to the criteria of autonomy, dynamic memory, and incremental learning. The rest falls more into what analysts call “agent washing”—a form of abusive rebranding of traditional tools.

Three Underestimated Structural Flaws

Unclear Business Value

In many cases, agents promise to automate complex tasks—but without demonstrating real economic relevance. Business leaders mention costly pilots, poorly integrated with existing systems, whose impact remains marginal. The alignment between the agent’s level of autonomy and business objectives is often insufficient, leading to strategic disillusionment.

Exponential Technical Debt

Implementing agents requires appropriate infrastructure: secure connectors, workflow orchestration, algorithmic supervision, data governance. However, many projects underestimate the integration effort and scaling costs. Result: agents struggle to leave the laboratory, or degrade the overall performance of existing systems.

Persistent Technological Limitations

Finally, algorithmically speaking, agents remain vulnerable: hallucinations, context errors, difficulty adapting behaviors to new or ambiguous situations. The LLM models on which these agents rely struggle to adapt to long-term objectives or effective feedback loops. Intelligence remains localized and opportunistic—far from generalized cognitive autonomy.

A Still Immature Market, but Rapidly Structuring

Despite this warning, Gartner does not dismiss the transformative potential of agents. The firm predicts that by 2028, nearly 15% of companies’ daily decisions will be made or suggested by AI agents, compared to 0% in 2024. Additionally, one-third of business applications could integrate an agentic dimension.

This projection, however, requires a strategic refocus: fewer promises, more rigor. Gartner recommends identifying high-impact, moderate-complexity use cases, avoiding hype around still-experimental solutions, and supporting IT and business teams in skill development. The agent becomes a strategic tool, but cannot be improvised.

Reintegrating Humans into the Decision-Making Loop

A key lesson emerges: Agentic AI does not eliminate the human factor—it reconfigures it. Agents must be part of supervised human value chains, where their contribution is understood, contextualized, and validated. Too many projects have failed due to lack of clear governance, prior ethical evaluation, or sufficient user buy-in.

The most advanced companies are now experimenting with collaborative multi-agent architectures, combining specialized agents with human validation committees. The challenge is no longer simply to delegate tasks to machines, but to redefine interactions in a hybrid, co-evolutionary system.

Toward Selective Maturity

In the short term, Agentic AI projects will concentrate in a few high-potential verticals: cybersecurity, legal assistance, financial workflow automation, document management, software development, and supply chain. Companies that succeed in extracting value from these agents share several common traits: rigorous management, robust infrastructure, deep process knowledge, and an incremental innovation strategy.

The message is clear: test, measure, learn, adjust. The era of technological “moonshots” without a solid roadmap is coming to an end. Agentic AI is not magic. It is a tool—powerful, but demanding.

Open Questions for AI Decision-Makers

  • What are the truly value-creating use cases in your organization?
  • Do you have an ethical and operational governance framework to oversee agent autonomy?
  • Does your current infrastructure enable smooth orchestration of interconnected intelligent agents?
  • How are you preparing employees to collaborate with semi-autonomous entities?

Gartner is not questioning the momentum of Agentic AI. It offers a lucid and beneficial perspective. Far from giving up, this prediction invites us to reassess our ambitions, resize our projects, and build step by step truly useful artificial autonomy. The agent will not replace humans—it requires humans to remain strategic.

 

Source:https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027