How to scale AI agents for business

When meeting with business leaders, there is excitement around the potential of what agentic AI can do for an organization. There is also a clear need to answer the question of how business leaders can effectively and efficiently deploy agentic AI.

New research from the IBM® Institute for Business Value shows the buy-in and excitement from business leaders: 86% of those surveyed expect process automation and workflow reinvention to be more effective with AI agents by 2027.

Traditional AI or automation tools are offering useful, yet still marginal, productivity gains but aren’t transforming the underlying process. With agentic AI, we can really start driving bigger and more strategic business outcomes that can create greater productivity and efficiency in an organization.

It's not just about AI telling us what to do—it's about AI starting to do it. We need to move beyond AI assistants and expand what's possible with AI agents that can execute and adapt processes under human supervision. This shift requires real reengineering of how work gets done, unlocking the kind of value business leaders genuinely want to achieve.

Already 76% of executives surveyed say that they are operating and delivering proof-of-concepts that enable autonomous automation of intelligent workflows through AI agents, according to the IBM Institute for Business Value.

Every client I’ve worked with wants us to have a deep understanding of agentic AI, a credible point of view and experience scaling agentic AI. And for good reasons. Agentic AI comes with a lot of promise and immense potential to transform your business, but with it come technical demands and the need for a cultural shift within an organization.

From my own experience, I’ve learned that the ‘how’ has become a prominent focus for clients and organizations. They are keen on seeing incredible results in cost saving, efficiency and productivity. The following are my insights on how to integrate this technology and scale it to great outcomes.

Specific areas where we’ve seen agentic AI work include customer service, procurement, finance and the whole IT process, but what we’re seeing in customer service specifically is a significant opportunity.

In fact, we have transformed contact centers that used traditional chatbots and automation tools by pivoting to an agentic approach. Our agentic conversational experience approach introduces a coordinated team of AI agents capable of handling a broader and more complex range of customer queries, instead of a single scripted assistant, like a chatbot. This helps realize significant efficiencies while operating with a foundation of defined guardrails to drive compliance and consistency.

What makes agentic AI more effective than traditional chatbots is its ability to operate holistically—not just following scripts, but dynamically coordinating actions, adapting to exceptions and continuously learning. Agents don’t work in a fixed sequence. They collaborate with each other and with humans to determine the most efficient way to resolve complex tasks in real time.

There are preemptive measures and preparations that must be taken to implement agentic AI effectively and efficiently before an organization can scale solutions and see improved outcomes.

The first thing is to identify an opportunity within your business. For example, let’s say I want my procurement function to be more efficient, and I want to get an agentic solution implemented. IBM developed a methodology for clients to formally assess whether an agentic solution provides added value and enhances the process or workflow.

Our agentic AI readiness assessment approach is based on:

• A structured assessment executed through a blend of process mining and LLM-powered process analysis
• A design to identify business processes best suited for agentic AI and autonomous transformation
• Five pillars to evaluate how well a process can be reengineered by using AI

The second part of the ‘how’ considers underlying enterprise architecture capabilities and identifies how architectures may need to evolve. This may mean thinking beyond traditional integration layers and establishing a modern architecture designed for autonomous, AI-driven workflows. Some of the necessary capabilities include:

• Multiagent orchestration and event-driven integration
• Centralized agent catalog and lifecycle management
• Agent memory and long-term context stores
• Modular, AI-ready data products
• Governance, observability and security layers tailored to AI agents

Data remains core to the successful deployment of AI and is a critical part of the conversation at the start. Our point of view at IBM is that this agentic AI application can deliver value only if you combine experience, process and data.

Managing structured and unstructured data, ensuring data quality and protecting data privacy are ongoing challenges. Yet, with the right strategies in place, businesses can harness the power of AI to drive transformation and future growth.

There are three core challenges to consider when preparing a business for AI transformation.

• Access to data: It is estimated that in 2022, 90% of data generated by enterprises were unstructured. Organizations need to access that data wherever it resides and unify it for their use case.
• Quality and intelligent data for real-time analytics and AI: Your AI is only as good as the data you input. Can you trust that data for your AI models? Is it of sufficient quality and how do you objectively evaluate the quality of your data? Answer these questions before deploying AI.
• Data security: Whether on-premises or multicloud, data security needs to extend to the entire landscape. Consider all data no matter where it is in motion and whether it’s structured data or unstructured data.

Another key factor clients must consider is strong change management. Specifically, clients must consider the people who need to adopt AI as part of their daily work.

A tangible example is from the HR transformation perspective, a use case where we really need to rethink the roles of people and consider where AI might be most valuable. Many of our clients in the HR function think about upskilling and reskilling employees whose roles are being reimagined.

Change management should be an integral part of any AI transformation. It isn’t just a technical implementation being done; it’s a holistic process that requires the client to consider the entire ecosystem that makes the business run smoothly, including technology, processes and people.

This shift with agentic AI isn’t just a change for employees and a reconfiguration of job functions. For example, at IBM, reimagining processes to create workflows where AI can be integrated to create a seamless optimization is what makes a transformation possible and helped enable IBM to drive 3.5B in productivity gains.

We have the tools and expertise in place to advise our clients on the right strategy and method to bring agentic AI into their business.

Once the ‘how’ has been established and a client understands what is necessary to scale agentic AI successfully, the next part of the process is to integrate agentic AI into the business.

1. Re-engineer for agentic: Agentic AI requires a shift in how work is designed and executed.
   Pro tip: Rethink workflows with agents. Delegate routine tasks to AI while elevating human roles for supervision, escalation and value-added judgment.
2. Ensure scalability: Scaling agents across systems and functions requires robust orchestration.
   Pro tip: Implement a strong agent orchestration layer that enables agents to work across platforms safely, coordinate tasks and respect process boundaries, and control layers.
3. Prepare your data: Agents need access to focused, high-quality data that’s actionable.
   Pro tip: Build use case-specific data products that are curated, governed and API-accessible to ensure that agents have the structured inputs they need to act in real time and in context.
4. Optimize for performance: Balancing speed, reliability and cost is critical at scale.
   Pro tip: Equip your platform to route agent tasks to the right LLMs and tools based on complexity and cost. Use caching, smart fallback models and usage controls to maximize ROI.
5. Test for reliability: Before deployment, agents should be monitored for fairness and explainability.
   Pro tip: Integrate agent evaluation into your AgentOps lifecycle. Automate testing for accuracy, bias, robustness and ethical compliance both predeployment and continuously in production.
6. Establish governance: Operational control and visibility are essential for trusted AI execution.
   Pro tip: Create a governance framework that includes observability, human-in-the-loop controls, KPI tracking and audit trails to monitor agent behavior and business impact.
7. Drive rapid deployment: Get to value quickly and build momentum.
   Pro tip: Start with high-value, narrow use cases that demonstrate impact fast. Use reusable agent templates and modular architectures to expand horizontally across functions.
8. Track business value: Impact must be tangible and measurable.
    Pro tip: Define KPIs such as workflow convergence, human handoff rates and business outcome improvements. Use them to guide iteration, adoption and executive buy-in.

When integrating agentic AI into your business, I have three recommendations:

• Reimagine processes: Don't just fix broken processes; rethink them entirely by using an augmented approach.
• Think beyond agents: Consider the entire end-to-end business process, including user experience, process orchestration and necessary data products. Think about the overall experience that you’re trying to deliver.

Plan for scalability: Design your AI architecture to scale quickly, starting with robust governance from the outset and quality data to work with right now and in the future.

Agentic AI is already at the center of enterprise innovation. Traditional SaaS platforms are evolving into agent marketplaces, where agentic apps can source, invoke and orchestrate AI agents across multiple systems to execute complete workflows.

Instead of relying on monolithic applications to perform rigid tasks, enterprises will begin deploying multiagent systems that dynamically coordinate work, adapt to context and reduce the need for manual intervention.

This transformation marks the beginning of a new architecture for digital operations—one built for autonomy, speed and continuous optimization.