Agentic AI has the ability to carry out complex tasks without human intervention. How scared should we be?
“Agentic AI is technology capable of autonomous decision making and action and it offers the bridge between personal productivity improvements and business-wide transformation,” says David Lee, chief technology officer at PwC.
At the moment, many businesses are struggling with how AI can deliver bottom-line returns – how to turn, say, time saved by AI into profits.
“Agentic AI offers a pathway for businesses to fundamentally reimagine their processes, moving from isolated task automation to organisational wide process transformation,” says Lee.
READ MORE
A recent PwC US survey of senior executives found 79 per cent were already adopting agentic AI in their companies. Of those, two-thirds said they were delivering measurable value through increased productivity.
“Even in functions that have already embraced AI, such as fraud detection in financial services, agentic systems offer a step change in capability – moving from identifying suspicious patterns to orchestrating the end-to-end investigation and resolution process,” says Lee.
“Similarly, in customer service, organisations are reporting not only reduced servicing costs but increased cross-selling and customer experience through AI-augmented interactions.”
Industry-specific applications are equally compelling. Healthcare providers can optimise appointment scheduling and diagnostic workflows, reducing wait times while improving resource utilisation. In consumer goods, content creation and campaign optimisation can be both accelerated and personalised at unprecedented scale.
“The most valuable lesson from early adopters is perhaps counterintuitive,” says Lee. “The strongest financial returns often come not from cost reduction through displacement but from capacity expansion through augmentation. In capacity-constrained organisations the ability to do more with existing resources represents the greatest win.”
Agentic AI works for people as well as organisations, says Ivan Jennings, senior manager in solution architecture at enterprise software company Red Hat.
“Let’s take a consumer example and imagine we give an agentic AI system the following instructions: ‘Find me a flight next weekend to Paris, I don’t want to spend more than €300; if you find a suitable flight, book it.’ The agentic AI will autonomously figure out the steps involved to reach that end goal with the constraints that we provided it with,” he explains.
“Let’s just digest that for a moment – we’ve told a system in plain text English, what we wanted to happen, when we wanted it to happen and what the maximum we would pay for it to happen is. If you switch the lens from a consumer to an enterprise business the use case list is long.”
He is seeing enterprises automating tasks that were once too complex, and require pulling data from disparate systems, such as all the interactions with a customer across different channels.
“Agentic AI can give a single pane in the glass view of where a business has engaged with a customer and assess the next steps to progress the relationship. Managers can use this information to understand how well the team is engaging with customers,” he explains.
Enterprises can see the benefits at scale. “We’ve seen agents interwoven with chatbots to provide consumers with the ability to get actual help for a multitude of issues. For example, we’ve all probably experienced trying to talk to someone from an ecommerce website only to be met with a chatbot. Previously, they lacked contextual awareness of a situation. With AI agents we can completely transform that customer experience,” says Jennings.
But there are risks too. “We might be as a society coming to terms with understanding the normal method for interacting with generative AI from the likes of Google Gemini and OpenAI ChatGPT in the form of a question and an AI response. With agentic AI we are asking it to do something for us. This is where a lot of hesitation comes in from enterprises around where they use agentic AI and where they don’t.”
For example, agentic AI unleashes the potential to bypass cybersecurity controls put in place to protect enterprises.
“There are ways to safeguard around this – using input and output guardrails is a great starting point,” says Jennings. “On top of this, consistent and constant observability is essential – ensuring that a team is viewing regularly what the agentic AI system is doing and what actions are being taken.”
It will inevitably impact on employment. “There will be changes in the workforce – there always are,” says Stephen Redmond, head of data analytics and AI at BearingPoint. “But if you look at Irish American organisation Intercom, for example, it specialises in customer service and has invested huge amounts in AI that they can deliver to their customers to be able to fulfil that customer service role, but they are not firing people.”
While Redmond’s hope that agentic AI may help us finally achieve the much vaunted four-day week may fall short, it will certainly ease recruitment challenges in a tight labour market, he points out.
There is, however, a risk that AI start-ups are too focused on the kind of things that humans love to do, rather than the tasks we don’t, he cautions: “I might want my AI to handle my calendar or to do lists for me. I don’t want it to do my strategic thinking for me.”
