Conference rooms are consumed by it, investors are wagering on it, decision makers are testing it, and Gartner analysts predict that by 2028, one-third of enterprise software will incorporate agentic AI — rising from merely 1% in 2024 — driving 15% of daily business decisions to be handled autonomously by then.

However, amidst all the excitement, something feels amiss, and most businesses remain caught in their trials, many of which never progress to full-scale implementation or falter during deployment. For perspective, 85% of AI initiatives fail. When you ask those developing these solutions what’s truly happening, the recurring theme is that although they possess AI agents, they lack the ecosystem to support them.

Constructing Infrastructure Initially

Aishwarya Singh, Senior Vice President of Digital Collaboration Services at NTT DATA, has witnessed this narrative unfold firsthand. “The main economic hurdles involve the substantial upfront investment in infrastructure and technology, the expense of merging AI with current systems, and the demand for specialized personnel to oversee and maintain AI systems,” she stated during an interview.

In theory, agentic AI should cut costs and simplify processes. Yet, in practice, it introduces additional layers of both — particularly if firms approach it as a product rather than a process. “Many leaders underestimate the time, effort, and resources necessary for successful integration,” Singh remarked. “Overlooking this can result in project delays, budget overruns, and subpar performance.”

Introduced in March of this year, NTT DATA’s new Agentic AI Services, developed using Microsoft’s CoPilot Studio and Azure AI Foundry, aim to resolve this issue—not just by deploying agents but by supporting the entire lifecycle: consultation, creation, execution, monitoring, retraining, and optimization. It’s AI infrastructure as a managed service, and it’s already being implemented internally throughout the company.

“In our internal ticketing systems, productivity increased by 50 to 65%,” Singh mentioned. “We develop agents across various ticket types and connect them across omnichannel LLMs so we can consistently add new automation through voice, email, and chat.”

The Shortage of AI Talent

However, the absence of infrastructure or ecosystem isn’t the sole factor hindering agentic AI. Another issue, potentially even larger, is the shortage of AI talent. According to a recent Accenture study involving 3,400 executives and 2,000 enterprise projects, only 13% of AI efforts yield significant business benefits. The reason? Businesses spend three times more on technology than on people—and this AI skills gap is evident.

“Readiness in talent is one of the biggest obstacles to scaling and extracting value for companies,” said Jack Azagury, Group Chief Executive for Consulting at Accenture. “You can invest in all the available Gen AI tools, but if your staff doesn’t know how or why to utilize them, the value simply won’t materialize.”

Singh concurs, highlighting that this widening AI talent gap is why NTT DATA is investing in training 200,000 employees and certifying 15,000 GenAI experts this year alone. “This has also sparked numerous ideas on how we can harness this technology to enhance our own business performance, leading to remarkable new innovations,” she added.

The AI Deployment Dilemma

Once you surpass the talent predicament, you encounter an even greater challenge in actually deploying AI. A recent working paper from the National Bureau of Economic Research examined AI chatbot usage across 7,000 workplaces and discovered that these chatbots had virtually no notable impact on wages or work hours in any profession. Despite widespread adoption, the study found that on average, AI only saved employees 3% of their time, with just 3 to 7% of that translating into higher compensation.

Even more striking is the finding that most employees redirected their saved time towards other tasks, often ones created by the AI system itself—editing AI output, verifying fabricated facts, or adjusting tone. In essence, the technology added more complexity than it removed.

That mirrors what IBM also uncovered in a separate study showing that only 25% of AI projects achieve their expected return on investment (ROI). And Informatica’s latest report reveals that data quality and integration problems remain the primary cause of most AI project failures.

The takeaway is that AI agents don’t expand because enterprises haven’t figured out—or understood—how to scale the surrounding conditions.

The Post-Deployment Complexity

If you manage to deploy your AI agents effectively, you now have to contend with what occurs afterward. Even the finest AI agent necessitates a team behind it: developers, data stewards, security architects, trainers, ethicists, and more. This is where most companies encounter the greatest difficulty, according to Singh—not in deploying an agent but in handling what follows.

“Post-deployment, agent management entails routine updates, performance tracking, security audits, and alignment with evolving business objectives,” she informed me. “A major concern we’re hearing from clients is how best to handle the influx of agentic AI agents within their organizations.” That’s precisely where many organizations are navigating blindly, constructing AI agents without a strategy for keeping them running, governed, and optimized at scale.

To tackle this mounting challenge, Singh mentioned that NTT DATA is beginning to introduce guardian agents and Red Teaming agents—models designed to monitor security, compliance, and operational integrity as agents proliferate across functions—into their managed stack.

Where Genuine ROI Starts

So what’s succeeding? If agentic AI is weighed down by all these complexities, why is there still so much enthusiasm about it, enough for many global companies to plan an agentic AI shift? Singh’s response is that despite the challenges and setbacks, agentic AI has practical applications that offer a glimpse of its potential when properly deployed.

“We’re observing top use cases in IT services, tactical process automation, customer service, and multi-agent models for more intricate tasks like inventory management,” Singh clarified. “Clients can anticipate a payback period of 6 to 12 months. Productivity gains typically become apparent within the first few months.”

Yet, those outcomes emerge solely when there’s a comprehensive system backing the agent—one that includes change management, talent development, cross-platform integration, and ongoing optimization. As Singh pointed out, companies that succeed are the ones who prototype swiftly with tactical use cases and hyperscaler-aligned teams prepared to scale within their existing cloud environments.

Mastering The Fundamentals

Agentic AI won’t expand because you hired a vendor. It will expand because you established the internal architecture—including technical, organizational, and human—to support it. That’s the key message for businesses planning to scale agentic AI today, according to analysts, projections, and several enterprise case studies.

Every successful agentic AI story begins with mastering the basics—data, talent, and infrastructure. And that, Singh emphasized, demands considerable planning. The question isn’t whether companies can scale their agentic AI projects. It’s whether they’re willing to do what it takes to make it happen.

The above is the detailed content of What It Really Takes To Scale Agentic AI. For more information, please follow other related articles on the PHP Chinese website!