Experts Highlight Marketing Aspect of 'Agentic AI' in Current Discourse, Urging Caution

Vadym Kazulkin, a prominent voice in the technology sector, recently asserted that the term "Agentic AI" is largely a marketing construct, applied to systems that are fundamentally "bounded, scripted automation with probabilistic components." This perspective, shared via a social media post, challenges the widespread perception of these systems possessing true open-world autonomy and underscores a critical distinction between advanced automation and genuine AI agency.

"Signal v. Noise: Agentic AI is marketing," Kazulkin stated in his tweet, elaborating that the term is being applied to "bounded, scripted automation with probabilistic components that appear proactive but lack open-world autonomy."

The concept of Agentic AI typically refers to systems designed to act autonomously to achieve goals, often involving planning and tool use. However, a growing chorus of critics, including Kazulkin, argues that many solutions marketed as "agentic" are sophisticated task orchestrators rather than truly autonomous entities. Experts like Gary Marcus and Yann LeCun emphasize that Large Language Models (LLMs), which underpin much of current "Agentic AI," lack intrinsic intentions or genuine decision-making, merely generating plausible continuations based on statistical patterns.

This distinction is crucial as it addresses a trend of "agent washing," where existing AI assistants and automation tools are rebranded without substantial agentic capabilities. Gartner predicts that over 40% of Agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. This highlights a disconnect between the marketing narrative and the practical realities of deploying these technologies at scale.

While current Agentic AI excels in narrow, well-bounded tasks with human oversight, its limitations include error compounding in multi-step workflows, high token costs for complex operations, and integration friction with legacy systems. The promise of fully autonomous general agents remains largely theoretical, as these systems struggle with real-world grounding, long-term memory, and reliable action execution without continuous human intervention.

For businesses, understanding this nuance is vital to avoid unrealistic expectations and potential liabilities. The most effective deployments of Agentic AI today are those that augment human capabilities, handling routine or data-intensive tasks while humans retain oversight for critical decision-making and exception handling. This approach emphasizes human-in-the-loop models, ensuring accountability and mitigating risks associated with truly unsupervised AI.