|
Getting your Trinity Audio player ready...
|
As generative AI reaches maturity and its limitations become more apparent, a new paradigm is emerging: agentic AI. Touted as the next evolution in artificial intelligence, agentic systems promise autonomy, iterative reasoning, and goal-oriented action. Yet, despite growing interest, the term remains loosely defined and frequently misunderstood—fueling both hype and skepticism.
A recent Gartner report predicts that 70% of agentic AI projects will fail by 2026. Critics argue this figure is inflated, pointing to widespread mislabeling of AI initiatives as “agentic.” So what truly defines agentic AI? Where does it fit in the broader AI landscape? And is its future as precarious as some suggest?
🧠 Defining Agentic AI
Reece Hayden, Principal Analyst at ABI Research, describes agentic AI as a system powered by large language models (LLMs) that can autonomously perform tasks, apply reasoning, and achieve outcomes—often through iterative processes. Unlike generative AI, which responds to prompts and produces content, agentic AI acts independently to fulfill objectives.
Hayden draws a compelling analogy: generative AI is akin to a recipe book, while agentic AI is the chef who takes an order and prepares the dish. This shift from passive response to active execution marks a fundamental transformation in AI capabilities.
Agentic systems vary in their degree of autonomy:
- Human-in-the-loop: A person reviews and approves each step.
- Human-on-the-loop: Oversight occurs at the beginning and end.
- Human-out-of-the-loop: Fully autonomous cycles, though still requiring safeguards against errors or hallucinations.
The core distinction lies in the AI’s ability to reason, act iteratively, and operate beyond prompt-response cycles.
🛠️ Applications and Limitations
Initial use cases for agentic AI are emerging in customer service and human resources. Here, agents interpret prompts, reason through data, and deliver verified responses—streamlining workflows and enhancing decision-making.
Hayden sees promise in these domains but cautions that broader adoption hinges on operational feasibility rather than technological readiness. “The challenge is not the tech—it’s regulation, governance, risk tolerance, and operational complexity,” he explains.
Looking ahead, Hayden envisions agentic AI enabling supply chain optimization, personalized retail experiences, and dynamic client engagement. For example, a manufacturer could use agentic AI to align supply with demand in real time. While such scenarios remain aspirational, Hayden argues they are technically achievable.
However, regulatory hurdles loom large. Removing humans from operational loops raises accountability concerns. “The real value lies in full autonomy,” Hayden notes, “but the feasibility of achieving that—even long-term—is questionable.”
⚖️ Skepticism and Misuse of the Term
Tom Cox, CEO of virtual sales agent firm 15Gifts, challenges Gartner’s pessimism. He argues that agentic AI is too nascent for such sweeping judgments and that the term is often misapplied to generative systems.
“True agentic AI makes decisions and fulfills tasks independently,” Cox explains. “It’s self-learning and self-improving—not just responding to prompts.” He warns that the term has become a catch-all for any advanced AI, diluting its meaning.
Cox also critiques the fear-driven investment strategies of many firms. Companies are expanding AI teams and budgets under pressure from shareholders, often without clear business cases. “The reality and cost of deploying agentic AI are immense,” he says. “Many initiatives fail due to poor data quality and unclear objectives.”
💸 Cost, Control, and the Path Forward
Agentic AI’s cost of ownership is a growing concern. Systems must be tailored, maintained, and updated for specific tasks—driving up expenses. Cox predicts that firms will begin scaling back AI investments as returns fall short of expectations.
He advocates for outcome-based approaches and cautions against premature adoption. “AI is rubbish in, rubbish out,” he says. “Without high-quality, structured data, the results will disappoint.”
One solution lies in partnering with specialist vendors, particularly in emerging markets. These vendors can deliver targeted agentic solutions on a pay-per-performance basis, reducing upfront risk. Operators can then orchestrate between in-house and external agents to build customized AI stacks—retaining control without bearing full responsibility.
🔍 Conclusion: A Pragmatic Frontier
While Gartner’s forecast may reflect the high failure rate of AI pilots, Hayden believes success depends on pragmatic implementation. Trialing agentic AI in real-world scenarios and investing in customer adoption are key to unlocking its potential.
Agentic AI represents a bold step forward—but one that demands clarity, discipline, and strategic foresight. As the sector matures, separating substance from hype will be essential to realizing its transformative promise.
