When generative AI systems such as ChatGPT emerged around three years ago, the business world was quick to adopt them as a means of improving productivity and reducing operational costs.
Now the latest evolution of AI technology is sparking even greater interest. Dubbed Agentic AI, it involves the introduction of self-directed software agents that can execute complex tasks without human intervention.
By leveraging data captured in multiple formats and iterative learning, the agents make informed decisions independently. Industry analysts project that, by 2028, 33% of enterprise software applications will incorporate Agentic AI, a significant increase from under 1% in 2024.
Challenges to be overcome
Despite this impressive forecast increase, significant barriers remain that threaten to slow the implementation of Agentic AI within organisations. The most pressing challenge lies in interoperability - or rather, its absence. Most AI agents today remain locked within proprietary technology stacks, with platform-specific memory and orchestration systems that prevent seamless co-ordination across different platforms.
This limitation means that what should be flexible, adaptive tools are forced to operate in technological silos. Without the ability to share context or delegate tasks across systems, these AI agents cannot fully realise their potential.
For organisations looking to scale AI adoption across different business units, this stalls AI deployment at the boundaries of vendor ecosystems, thereby reducing the business benefits that can be achieved.
When it comes to consumer-facing applications, trust has also emerged as a critical concern. For example, if AI agents make purchase decisions on behalf of shoppers, trust needs to be established that the transactions will be correct.
However, like most Large Language Model (LLM)-powered applications, agentic AI may operate in 'black boxes', where its internal decision-making processes remain opaque to users or even developers. This lack of transparency can erode consumer confidence, particularly in situations where personal and financial information is at stake.
The role of knowledge graphs
Addressing these twin challenges requires a fundamental shift in the way the data that powers AI agents is structured and connected. Knowledge graphs have emerged as a critical technology for bridging these gaps by providing a shared data foundation that enables agents to reason over connected, context-rich information.
The unique structure of knowledge graphs, made up of 'nodes' representing entities and 'edges' showing the relationships between them, creates a framework suited for AI reasoning. This approach solves the interoperability problem by establishing a common structure that different agents can use. It also establishes a standardised repository for AI agents to access information, regardless of their underlying platform or vendor.
Trust also improves substantially as graph-based systems offer transparent data lineage and clear reasoning paths. AI decisions become explainable through linked context, which also grounds AI outputs in factual and context-rich data, reducing the 'hallucinations' that plague LLMs when they operate with unstructured or inaccurate data.
Achieving comprehensive insights
What users discover with GraphRAG (Retrieval Augmented Generation using knowledge graphs) is that responses become not just more accurate, but richer, more complete, and consequently more useful. GraphRAG applications incorporate knowledge graphs in the information retrieval process for AI models, enabling AI agents to make informed decisions.
Consider a practical retail scenario. An AI agent addressing a customer service query could recommend a personalised discount package based on a comprehensive understanding of the customer, drawing on connected information about their tenure, current service usage, and history of interactions.
Without this connected view, an agent might offer inappropriate discounts based on fragmented data, creating both customer confusion and business losses. However, with a knowledge graph connecting disparate pieces of information, the agent can see that a customer has been loyal for five years, currently uses three distinct services, and has recently filed a complaint, enabling truly personalised and appropriate recommendations.
Laying the groundwork for Agentic AI
The performance of Agentic AI hinges on the quality of the data that powers it, and the integration of it with knowledge graphs is therefore a game-changer. With connected data providing the deeper context for agents to reason effectively, they will deliver greater real-world impact with smarter results.
Businesses need to recognise that the LLMs and AI agents are evolving rapidly with each iteration. Agentic frameworks are steadily lowering the barriers to building sophisticated, multi-step applications that can transform operations across all sectors of business.
Australian organisations need to take the steps now to open up their data stores and make them AI agent-friendly. Doing this will deliver significant benefits in the months and years ahead.
