Insights

Agentic AI - tapping into autonomous value creation

Adnan Masood, PhD, Chief AI architect, UST.

With chain-of-thought reasoning in DeepSeek making headlines, the dynamic reach of OpenAI Operator, and the advanced capabilities of Claude “Computer Use”, agentic AI is now top of mind for executives everywhere– emerging as the ultimate strategic imperative. Freed from static prompts, these agentic systems autonomously plan, execute, and adapt. In a business world driven by speed and personalization, they deliver breakthroughs once reserved for human teams alone.

Let’s dig in!

DIVIDER

A gentle introduction to Agentic AI

Agentic AI refers to artificial intelligence that can set goals, plan actions, and carry them out—all with minimal human supervision. Instead of merely producing answers, it actively solves problems, adapts to new information, and completes tasks end-to-end.

Agentic AI goes further than traditional automation by granting software the authority to learn, decide, and act on business objectives. Instead of providing answers alone, these agents manage tasks, coordinate data flows, and even resolve unforeseen challenges. This jump from reactive to proactive AI is a radical shift in how we lead teams and allocate capital. My experience shows that adopting Agentic AI fosters a culture of experimentation, elevates accountability, and transforms how we view technology investments.

DIVIDER

Agentic AI in the industry

As a cross-cutting enabler, Agentic AI is not confined to a single industry vertical. For example, in finance, several instances of specialized AI agents can help automate compliance checks, while others oversee ledger integrity, and provides on-demand risk insights. In the healthcare industry, we are witnessing a rise in virtual care assistants scheduling appointments and monitoring vitals in real time. Retailers use AI agents to forecast stock, handle personalized promotions, and power self-driving shopping carts. Manufacturing and IT operations tap into AI’s steady workflow orchestration for everything from machine health to DevOps pipelines. There is a unifying theme: routine tasks vanish while creative and strategic capacity soars.

But business leaders want measurable outcomes, not just tech buzzwords or art of the possible.

We see agentic AI making many promises-to deliver quantifiable gains by cutting tedious steps, freeing time for complex human judgment, and curtailing mistakes. It can compress project timelines and reduce headcount costs for repetitive functions. Then come the revenue drivers—enhanced products, new monetization avenues, and faster market pivots. The result is a balanced sheet reflecting capital efficiency and dynamic growth.

But then the ultimate question becomes, can it deliver on these promises? No one wants just theoretical potential—are enterprises actually leveraging Agentic AI? There is definitely a forward-moving momentum in the industry – and the practitioners, experts and advisories agree. Gartner predicts that by 2028, a notable slice of daily decisions will be fully agent-driven. Forrester sees tens of thousands of workflows ready for this shift, driving new levels of cost optimization. However, it's still in the early stages of a broader adaptation. As the Agentic AI technology evolves and matures, many organizations I work with are running internal proofs-of-concept and limited-scale pilots. Leaders often keep these trials under wraps because of the inherent risks and complexities. Yet, we see small pilots transitioning into production, albeit with less external fanfare. Things are moving fast, and Agentic AI is rapidly maturing behind the scenes.

A large fintech client I closely work with is in the process of actively deploying a conversational compliance agent for its payment-processing platform. This agentic system—powered by Crew AI and LangGraph—scours the end-to-end document pipeline, ingests regulatory documents, annotates key requirements, and integrates with internal rule engines. It then dynamically creates and updates rules for the payment processor side, running anomaly checks on incoming data and triaging escalations without micromanagement. The agent frees compliance teams to focus on outliers and strategic decision-making by tying directly into backend systems. Long-term ROI is still to be seen, but the results are a leaner, more adaptive compliance workflow and a faster path to addressing hidden risks.

There are various other potential early adapter success stories, such as a major healthcare payor finalizing an MVP to trim admission errors through automated triage agents. Online retailers are testing real-time agent-driven cross-selling to increase average order values. Each success resonates in real numbers: fewer escalations, lower transactional costs, and elevated customer loyalty.

These promises and success stories come with a caveat, though - deploying agentic AI must come with strong data governance and alignment with management’s risk profile. McKinsey urges boards to incorporate AI oversight into their strategic planning cycles. As a practitioner, I echo their caution. But the consensus is clear: ignoring these capabilities leaves money and opportunity on the table.

DIVIDER

Implementation roadmap – How to get started?

My advice is to start small, but not trivial – let me explain. Select a high-impact but relatively controlled use case. Begin with a single, high-impact workflow you fully understand, like a routine compliance check or an onboarding sequence. Ensure you have clean, structured data to feed the agent and define clear success metrics—lower error rates, shorter cycle times, or fewer handoffs. Pick a proven framework (e.g., LangChain) or a trusted hyperscaler’s AI platform (like AWS Bedrock, Google’s Vertex AI, or Microsoft’s Azure Agent Service). Stand up a small pilot team: include a data engineer, a domain expert, and someone who understands security requirements. Build and test the agent in a sandbox environment before production, set up permission controls, and assign a clear “owner” for the rollout. Limiting the scope reduces risk and gains a fast win to build momentum. Don’t forget to engage your technology partners or internal architects early to ensure the right frameworks and cloud integrations are in place. Structure a governance model: define where human oversight is mandatory, how success is measured, and how changes in the agent’s autonomy are approved. This approach keeps the investment rational, showcases success quickly, and builds internal confidence.

DIVIDER

Frameworks and libraries for Agentic AI: Open Source, AWS, Google Cloud, and Microsoft Azure

Building AI agents from scratch can be daunting –you shouldn’t do this unless that is part of your core business. I’ve seen global enterprises achieve quicker wins by leveraging hyperscaler and open source toolkits. For example, AWS’s Agents for Bedrock expedite orchestration and integrate securely with core systems. Google Cloud’s Vertex AI Agents blend advanced natural language processing with robust search capabilities. Microsoft Azure’s AI Agent Service ties elegantly into widespread productivity apps—an attractive benefit for organizations already standardized on Microsoft. Each platform promises low entry costs, strong governance, and scalable infrastructure, so your decision typically hinges on existing vendor relationships and data compliance needs.

And then there are open source tools - Crew AI, LangGraph, AutoGPT, Semantic Kernel AgentFlow, and Flowise, to name a few - I recall a project that floundered until the team discovered LangGraph - Suddenly, they could link large language models to databases and user-facing tasks seamlessly. AutoGPT is also a great example of maintaining an ongoing decision loop without constant human prompts. OpenAI’s function calling provides an elegant method to invoke external services while remaining and remain flexible. Other tools like Semantic Kernel, IBM Watson Orchestrate, or specialized vector databases ensure these agents remain robust, interpretable, and easy to manage at scale.

But how do you choose an accelerated development pathway with fewer integration headaches?

Choose a platform that aligns with your vendor ecosystem, data compliance needs, and team expertise. Then leverage frameworks like LangChain, AutoGPT, or function calling to accelerate development and minimize integration pain.

DIVIDER

Challenges and opportunities

Like any emerging technology, Agentic AI faces multiple hurdles. First, subpar data quality may undermine decision accuracy, forcing leaders to invest in robust governance. Security concerns also intensify when software acts on its own, making authentication and access control paramount. Then there’s the issue of trust—employees and managers can be uneasy about an AI “deciding” policy matters, so tight oversight and well-defined escalation paths are a must. Integrating agents across legacy systems can further complicate deployments, fueling both technical and organizational friction.

Autonomous systems raise ethical and regulatory questions and unnoticed biases or misaligned objectives can lead to reputational and compliance risks.

To combat these issues, leaders enforce data governance, implement robust security, maintain human oversight, streamline integration paths, and embed ethical checks throughout each Agentic AI deployment.

DIVIDER
 

Future outlook and key takeaways

Agentic AI has shifted from a concept to a practical toolkit that can transform core business processes. It unburdens people from routine steps and raises the bar on operational discipline. With the right strategy, these agents have the potential to reshape your enterprise’s cost structure and enhance revenue. Leaders who seize this moment can achieve a market advantage rooted in agile execution and data-driven insights.

I have witnessed countless corporate initiatives stall when they fail to leverage the power of autonomy - Agentic AI bridges that gap. It allows our organizations to automate beyond static workflows and embark on true strategic transformation. operational overhead is reduced, and execution speed is enhanced. The immediate benefit is sharper financial control and healthier margins. In a fast-moving landscape, an autonomous digital workforce has become a realistic and compelling growth lever.

So what does the future hold? Boardroom discussions have moved beyond cost savings to focus on driving top-line growth—shaping the future of business strategy. Specialized AI agents emerge to make this possible —some focusing on finance tasks like forecasting budgets, others on marketing campaigns, and still others on resource allocation in manufacturing. We will also see robust agent collaborations, with multiple AI entities coordinating in real time. There will be new roles managing them and new regulations guiding them. This is not hype; it’s a sober assessment of how technology evolves when efficiency meets autonomy.

As we refine our methods, the real winners will be those who position agentic AI not as a side project but as a central pillar of their operating model.

Connect with me to learn more about leveraging Agentic AI for the enterprise