Agentic AI: an introduction

There has been increasing talk over the last year or so about agentic AI. Here at AIDA we like to think we know a little bit about it, so thought we would shine a little light on what it is, where it’s come from and where it’s going, together with some practical examples.

What is agentic AI?

Definitions vary somewhat. Anthropic defines agents as “AI systems which can autonomously plan and execute complex tasks.” Deep Analysis puts forward a slightly more technical definition: “Agents are software entities that interact with human or machine requesters to collect, reiterate and ultimately return information” although its analyst brief on AI agents also emphasises the importance of planning and execution.

This differentiates it from its predecessor: AI assistants, which most of us are probably familiar with in the guise of those first generation chatbots which were more limited, using simple keyword triggers to give responses. 

Unlike traditional AI that operates within narrow task boundaries, agentic AI systems are designed to operate as autonomous agents capable of setting goals, making decisions, learning from feedback and adapting to new environments with minimal or no human oversight. In other words, agentic AI refers to systems that operate as intelligent agents, meaning they can:

• Set objectives and plan actions independently
• React dynamically to changes in their environment
• Collaborate with other agents or humans
• Learn from past experiences to improve future performance

These agents are often powered by large language models (LLMs), reinforcement learning and multi-modal inputs that allow them to interpret and act on text, images and structured data. What particularly separates agents from previous generations of “chatbots” is the use of a ‘reasoning engine’ to determine a plan dynamically.  

Practical applications today…

Agentic AI is already being used across industries to tackle challenges that previously required human judgement and adaptability such as:

• The latest generation of customer service chatbots that autonomously escalate issues, rewrite responses based on tone or loop in the correct department.
• Financial assistants that proactively analyse market trends and suggest investment strategies.
• Logistics agents that re-route supply chains in real time due to weather, pricing or inventory changes.
• Order processing agents where an order needs to be changed, cancelled or even processed with the associated updating of systems and sending of associated communications, e.g. e-mails with order confirmations, invoices, credit notes or even a meeting appointment.

…and in the future

In the near future, agentic AI will start to spread its tentacles further into business operations, linking systems, departments and even organisations, automating processes between them all, as one AI agent starts to communicate with another. This will undoubtedly lead to a further evolution in the application of AI agents including:

• Autonomous business operations: AI agents coordinating tasks across departments, systems and stakeholders.
• Legal and compliance advisors: self-updating agents that interpret regulations and apply them to documents or contracts.
• Knowledge graph expansion: agents continuously updating enterprise knowledge bases by reading, summarising and linking documents dynamically.

Agentic AI and AIDA

At AIDA, where we focus on revolutionising document workflows through end-to-end automation and business intelligence, the emergence of agentic AI promises to reshape how tasks are performed, not just by automating them, but by enabling systems to think and act in complex, dynamic scenarios.

AIDA has long since built on its solid original platform foundation of Optical Character Recognition (OCR) and Intelligent Document Processing (IDP) to add capabilities around payment, business intelligence, document relations and most notably AIDA-GPT, which enables you to interact with your documents like you would with a colleague, asking for summaries, compliance flags or related documents in plain language.

There will certainly be more to come from AIDA-GPT. Imagine being able to ask AIDA if all your invoices have been sent out. Imagine AIDA creating all the missing invoices AND sending them out via email to your clients. Imagine if all that process was automated. Watch this space!