What Is Agentic AI?

Agentic AI is made up of autonomous agents that can make decisions and interact independently to complete objectives. These agents can be thought of as small digital assistants, each with specific capabilities, that choose the best way to complete a task without needing instructions for every step of the process…

Traditional software follows pre-programmed sequences and agentic AI is different because it can work more flexibly. If asked to process an order, for example, an agentic AI system might check stock, compare prices, manage back-orders, and offer delivery options without being told exactly how to do each action. This is possible because the system uses large language models to decide which capabilities to use and in what order.

Legacy agent systems have existed for years, but they lacked this autonomy. Their actions were fixed and predictable, making them useful for simulations but far less adaptable. The newer approach combines natural language instructions with tool-using capabilities, allowing agents to expand their functions over time and create their own execution plans.

Capabilities within an agentic AI system can be compared to building blocks, where each block is described in natural language so the system knows when and how to use it. The system can orchestrate actions in a set order, or choreograph them more loosely in response to events. For example, during order fulfilment, it might automatically send a customer update when an out of stock item becomes available.

 

How Is It Being Used In The UK?

 

According to Salesforce research conducted with NewtonX, 78% of senior business leaders in the UK and Ireland say their organisations already use AI agents in their workflows. The survey of 110 C-suite executives found most of these leaders see the technology as a core part of productivity improvement. Many companies report saving between three and ten hours per week through AI agents.

The same research shows 62% rate their technical infrastructure as good or very good for supporting AI, while 47% say employees are adapting well to using AI tools. Around 26% expect AI agents to boost productivity further across their business.

PwC’s 28th CEO Survey adds that 61% of UK chief executives are investing in AI, with agentic AI seen as a way to reimagine business processes. These autonomous agents are already being used for database updates, customer communication, and managing internal systems. The survey points to growing interest in scaling agentic AI across more complex, multi-stage business operations.

Organisations that have adopted agentic AI are also considering how it changes workplace structure. Salesforce found that 81% of leaders believe AI agents will transform organisational structure, and 91% think it will allow employees to move into new roles that complement the technology.

 

 

What Does This Mean For The Workforce?

 

With agentic AI becoming more common, business leaders expect changes in which departments grow. The Salesforce study reports that 65% expect IT teams to expand, followed by 62% anticipating growth in research and development, and 41% in sales. Customer service and operations roles are more likely to contract as agents take over repetitive tasks.

Leaders believe human and AI agent collaboration will become the most valuable skill, with 86% naming it as a priority. AI literacy is next at 84%, followed closely by adaptability at 83%. Creative thinking, problem-solving, and accountability in AI use are also rising in value.

To prepare, 84% of leaders say they plan to train employees on AI within the next year. On average, they expect to invest 12% of an employee’s salary in retraining for AI-related skills. Only 8% do not expect to invest in reskilling at all.

 

What Are The Risks Of This AI?

 

While agentic AI offers adaptability, there are also risks to consider. One concern is full autonomy, where systems make decisions without human oversight. Without checks in place, AI agents could act in ways that conflict with company goals or ethical standards.

Another issue is premature deployment. Moving prototypes into production without proper testing can lead to failures, especially with complex agent systems. These require careful design and clear architecture to avoid technical debt, where unplanned additions create messy and systems that are hard to maintain.

The technology also faces challenges from fast-changing frameworks. Frequent updates can break compatibility, requiring constant version management. Clear and precise language in describing system functions is essential, as vague descriptions can cause the AI to use the wrong tools.

Also, many systems use external large language models, which process text through tokens. Over repeated testing and development cycles, these token costs can add up. Switching between locally hosted and remote models without thorough testing can lead to inconsistent results.

Agentic AI in the UK is already past the early curiosity stage. Salesforce’s data shows active executive discussions in 90% of organisations, with 89% agreeing that integrating AI into workflows will be a central part of their work. This suggests that AI agents are becoming embedded in business strategy rather than treated as experimental projects.

PwC mentioned that while the technology can reimagine business models, success depends on careful engineering and strong operational processes. Testing, monitoring, and profiling remain as important as in traditional software.

If adoption continues at the current pace, the UK could see more businesses moving routine work to AI agents while retraining staff for higher value tasks. This could lead to more efficient operations, faster decision making and more flexible service delivery.