AI agents can autonomously make decisions and take action to deliver on specific objectives. Are they the next major step in entirely transforming how organisations operate?

According to IBM's latest report 'Agentic AI's strategic ascent', 24% of executives say AI agents are taking independent action in their organizations and by 2027 that number is likely to rise to 67%.

But 78% believe that to maximize agentic AI's full potential, businesses need to use it to design entirely new ways of working. The research identifies a clear divide between those using AI agents to achieve incremental efficiency in their existing processes and a group of forward-looking enterprises using it to rethink their operating models to drive 'net new capabilities'.

Based on a survey of 800 C-suite executives in 20 countries across 19 industries, the report explores how agentic AI is reshaping the next phase of enterprise transformation. Here are five key takeaways.

1. Incremental AI is no longer enough

78% of companies admit that most AI spend is focused on improving existing processes. This is to be expected, of course, because AI has the power to streamline and optimize existing activities and workflows. But the bigger opportunity is using AI to create "net-new business capabilities", the report suggests.

It identifies two types of organizations: process-focused and transformative.

The process-focused organizations are successfully achieving measurable business impact by using agentic AI to improve existing operations. But as AI becomes commoditized, their first-mover advantage from doing this will only get smaller.

Transformative organizations, on the other hand, are using agents to do entirely new things more efficiently, not just improving current processes. They're focusing on generating net-new capabilities by reimagining business operations using AI agents for autonomous decision-making.

In a regulated industry, the process-focused company would increase efficiency by using agentic AI to quickly identify and interpret global risks and regulatory changes, for example. The transformational one will use teams of AI agents to scan, interpret, and act on regulatory changes across different regions worldwide - and those agents will work in tandem with other AI agents to update communications, documentation, and approval processes in real time. The transformative firm will be able to mitigate risks while freeing compliance teams from manual work, enabling it to operate across regions with confidence that it remains compliant.

2. Transformative organizations ask different questions

One key difference between process-focused organizations and those in the transformative group is that the latter ask different questions about what's possible with AI. Which, interestingly, also means that the KPIs they use to measure AI are different.

Hence, the research suggests that transformational organizations are more likely to have developed new KPIs to measure agentic AI's impact than process-focused firms. The underlying message is: if you keep measuring AI using the same metrics you've always used for human-led processes, you'll limit its potential to transform how work gets done.

The old approach was based on measuring human activity and efficiency in a world where humans are the primary actors. But agentic AI requires a new set of KPIs that are focused on monitoring the outcomes of automated AI decision-making.

So the starting point should be - ask questions such as "What new value can we create with autonomous systems that we couldn't before?" For example, instead of just measuring productivity, you could start measuring areas such as new value creation, business growth and the velocity of innovation.

Specific examples of new agentic AI-related KPIs the report mentions include "agent-to-human handoff rates" (to help understand where agents are failing to solve problems autonomously) and "decision accuracy rates" (which assess the reliability and accuracy of autonomous actions).

3. Workforce readiness and human critical thinking are critical success factors

The success of agentic AI is closely tied to the readiness and capabilities of the workforce that will support it.

47% of executives believe that inadequate employee skills are a barrier to success with agentic AI. Hence, they'll need to recruit and train employees in entirely new capabilities such as AI workflow and experience orchestration, AI cybersecurity protocols and monitoring and multi-agent systems architecture.

At the same time, 79% believe "human critical thinking" - the ability to challenge or refine AI-driven decisions (and to understand and explain why) - will become a key way to differentiate their business.

Underpinning all this is the recognition that companies will need effective change management strategies that clearly communicate how employee roles will evolve alongside AI rollouts. How do employees fit into new AI workflows? And how will new opportunities be supported by upskilling and reskilling?

4. Tackling the AI trust deficit is all-important

45% of executives believe that a lack of visibility about how AI agents make decisions is a significant barrier to success, recognizing that autonomous AI agents need to 'show their workings'. Without this, how can the business - and its customers and partners - ever trust AI?

Businesses should aim to create AI that's transparent, measurable and continuously improving. As part of this, they need to find ways to address the AI "black box" so that every automated decision or transaction is audited and understood.

The report emphasises the importance of techniques such as machine learning operations (MLops) to continuously monitor the health of AI models, AI observability platforms to log and track every AI decision, and A/B testing to compare the outcomes of different AI-driven approaches.

The AI trust deficit further underlines the importance of recruiting talent to oversee and monitor AI and of providing training and support for new roles to do so. Winning the trust in AI becomes even more essential for transformative enterprises. The stakes become even higher if you're making strategic changes that redesign your business model.

5. These three factors make AI success 32x more likely

Companies that excel in three important areas are 32x more likely to achieve top-tier business performance using AI:

• Cybersecurity integration: When AI makes autonomous decisions, cybersecurity becomes even more critical. AI agents don't just make decisions based on data, they act on it. A malicious actor who can hack into a system to take control of an organization's AI agents can wield the power to do more harm than ever. Robust cybersecurity is the difference between agentic AI delivering a powerful competitive advantage or becoming an "existential threat".
• Ethics analysis in AI systems: Autonomous AI systems must be prevented from making decisions that are biased, wrong, illogical, or go against human values. Embedding ethical oversight and guardrails that ensure agentic AI is fair, transparent and accountable becomes mission critical.
• Workflow-specific small language models: It's better to use tailored small language models that understand the complexities, context and terminology of industry specific workflows. They are going to be much more effective and accurate at making the right decisions in the "messy, nuanced world of actual enterprise operations" than larger generic AI models.

According to IBM's report, agentic AI is more than another technology upgrade that optimizes and automates how things are currently being done. It's an opportunity to rethink how humans and autonomous systems can work together in entirely new ways to transform business operations.

This blog was originally published on the IBM Community.