According to Gartner’s forecasts, more than 40% of the projects involving agentic artificial intelligence will be abandoned by 2027. In question: poorly defined costs, unclear commercial value and a lack of risk control. Despite this, companies continue to invest in this technology.
The Gartner analysis firm estimates that by the end of 2027, more than 40% of the projects linked to the agency AI will be abandoned. Among the main factors of this failure are the increase in costs, uncertainty about the return on investment and gaps in risk management. The majority of projects are still only at the test of test or proof of concept. Many companies underestimate the effort and complexity that agency AI represents, which harms the concretization of projects.
A survey conducted in January 2025 with 3,412 participants in webinaries also shows that only 19% of companies have already invested in agenic AI. 42% prefer prudent investments, 8% do not plan to invest, while 31% remain undecided.
Another problem: the phenomenon of the “Washing agent”, which consists in presenting as agentic AI of applications that are not, such as assistants or RPA tools (Robotic Process Automation). According to Gartner, only some 130 suppliers now offer real agent solutions. But these models are often not yet mature enough to achieve complex objectives independently or execute nuanced instructions over time.
Growth expected despite everything
Despite these cautious perspectives, Gartner anticipates that on 2028, at least 15% of daily professional decisions will be taken by IA agents. However, their deployment will have to remain progressive and thoughtful. The firm recommends using them only in cases where their commercial value is clearly identifiable. Integration into existing systems remains complex and requires deep adaptations of the processes.
Companies are invited to focus on productivity gains, based on AI agents to automate repetitive tasks and facilitate decision -making. The objective: to reduce costs, while improving the quality, speed and scalability of operations.
