Consider this a quick reference guide for what this latest form of business-oriented AI brings to the table.
Agentic AI has quickly established itself as the next evolution of artificial intelligence for business.
This puts it on a similar growth trajectory to the early days of Generative AI, and with it come some of the same initial questions: 'What's it good at?' 'How is it best used?' and 'What's in it for me?'
At its heart, Agentic AI is about creating AI agents that can act as intelligent digital workers. The intelligence comes from AI language models, which give the agent an encyclopaedic knowledge of different approaches to a situation. You just need to guide it on which resolutions are acceptable.
Specific instructions constrain the AI agent's ability to act and respond, likely by additional human-in-the-loop oversight at first, which can be wound back over time once the agent proves capable of doing what it's been assigned.
Key strengths
A key difference between Agentic AI and technologies previously used to automate the execution of parts of a process is the expansive range of tasks that can be performed.
Existing technologies, such as robotic process automation (RPA), helped organisations realise process efficiencies. However, they were mainly suited to repetitive, high-volume, low-value tasks like verifying a customer's identity, checking order status, updating account information, or sending confirmation emails.
Agentic AI has three key strengths: its ability to handle unstructured information, its adaptability when encountering exceptions or issues, and its ability to improve a process's overall reliability.
Collectively, these strengths open the door for businesses to do far more with process automation than before, with many more processes now also becoming candidates for improvement.
It's worth exploring these strengths in more detail, as they will likely be critical to the business case for piloting or producing Agentic AI in 2025.
Strength #1: Working with unstructured inputs
Many processes start with an email. For the internal finance department, this could be an invoice emailed to accounts payable or a supplier asking when a submitted invoice will be paid.
RPA could help by monitoring the central inbox, downloading an invoice attached to an email, and extracting key details from it to populate into the finance system automatically. This assumes the email, details, and formatting of the invoice will remain consistent and recognisable over time.
In reality, emails and documents people produce vary in structure, content, and the nature of assistance sought. For example, a supplier may file an invoice but also ask about the status of an existing invoice. They don't write their email with an RPA bot in mind as the intended recipient; they expect a personalised response from a finance team member. The RPA that triages it isn't capable of handling the ambiguity of the multi-topic email and has to hand the entire request off.
Agentic AI approaches things differently. It still monitors the central inbox for emails with an invoice attached. Still, it can understand the unstructured nature of the supplier communication, get to the root of what the email is asking, and take action. Depending on its design, Agentic AI may send the invoice to an RPA workflow to be processed and then poll the ERP or finance system to find out the payment status of the earlier invoice. It might then draft an email with the information back to the supplier.
Strength #2: Adaptability when things don't go to plan
A hallmark of Agentic AI is adaptive workflow management, enabling agents to adjust processes in real-time based on changing conditions.
Imagine an ERP system outage invoice processing. Instead of grinding to a halt, the AI agent adjusts. It might switch to a backup workflow, use local data sources, or notify key stakeholders about the delay. Or perhaps during processing, the AI agent encounters an invoice that doesn't have a corresponding purchase order in the ERP. At this point, traditional automation would likely stop or fail. Instead, the AI agent recognises the missing PO as an exception and flags the invoice for further investigation. It checks supplier history, looks for similar transactions, and cross-references other internal systems to validate the invoice.
This adaptability ensures resilience and efficiency, even in unpredictable scenarios.
Strength #3: More reliable processes
If one AI agent in a single process can be highly effective, the impact expands when multiple interconnected processes are each augmented by Agentic AI.
We anticipate seeing AI agents coordinating with each other to achieve broader business goals within three years. Behind-the-scenes orchestration will be required to ensure this works smoothly, but it will enable automation of complex multi-stage processes, including managing hand-offs between different functional areas of the business. On the customer side, this will mean expedited service delivery interactions and outcomes.
More AI agents in processes will also increase the reliability of end-to-end processing since more parts are automatable, making outcomes more predictable.
