Agentic AI, one of the newest concepts in artificial intelligence is now gaining real traction. It has gone beyond its initial buzz. Agentic AI is accelerating the development and deployment of autonomous business systems by building on machine learning.
This technology, which operates as an independent “agent”is able to make informed decisions using multimodal data, algorithmic logic and can then “learn” and evolve through experience.
What’s even more exciting is that it can act independently. Agentic AI is distinguished from previous generations of AI tools by its unique ability to plan, adapt and execute complex tasks without human supervision.
AI agents in supply chains can track market activity, historical demand trends, and implement measures to prevent shortages. For example, by automating certain restocking processes. These agents adapt their behavior to changing market conditions and improve efficiency and performance. It’s no surprise, then, that 26% of leaders in business report that their organizations are forming strategic approaches around Agentic AI.
As great as it may sound to outsource these tasks to Agentic AI we should also err on caution. How can AI agents’ actions and outputs be trusted, despite their autonomy? How can we trust Agentic AI to perform complex tasks on its behalf? Or how can we be sure that its decisions are based on what is happening in the real-world or the enterprise’s perspective of the world.
Just as our brains rely on observation and additional inputs to draw conclusions to enhance their reasoning abilities, AI agents must rely on many external sources and signals.
Platforms and solutions that collect and present data easily accessible and retrievable can meet this need. Here’s how:
The trust challenge for autonomous AI systems
It has been discussed that what sets Agentic AI aside from other AI systems, is its ability to act independently, and not just engage in linear conversations. Agents are often required to consult multiple dynamic external sources to complete complex tasks. The risk of something going awry increases as a result. You might trust a chatbot for an update on a refund or claim, but would you trust an AI agent to book your flight using your credit card?
Task-based agents are different from conversational AI. They plan and change their actions based on the context that they’re given. They delegate subtasks through a process known as “chaining”where the output of one action is used to input the next. This means that queries can be broken into smaller tasks. Each task requires real-time access to data, and is processed iteratively in order to mimic human problem solving.
In order to make decisions, the chain effect is influenced by the environment being monitored. For two reasons, it is important to have accurate and explainable data retrieval at each stage of the chain. Users need to understand why the AI agent made a certain decision and the data that was used.
Users need to trust that an action is effective and efficient. Second, they must be able optimize the process in order to get the best result every time. They do this by analysing the output at each stage and learning from unsatisfactory outcomes.
The value of data is multiplied by a large amount when you trust an agent to perform complex tasks based on several retrieval steps.
It is important to make enterprise data reliable and available to agents. Businesses are increasingly recognizing the power of graph databases for the wide range of retrieval options they offer, which in turn increases the value of data.
How graph technology enhances AI reasoning
Agentic AI makes decisions based on data. The insights that support these decisions need to be accurate, transparent and understandable. Graph databases are uniquely suited to deliver these benefits. Gartner has already identified knowledge graphs as a key capability for GenAI applications. GraphRAG, where the retrieval path contains a knowledge network, can improve the accuracy of the outputs.
Knowledge graphs are unique in their structure, which is made up of ‘nodes and edges’. This allows for higher-quality answers to be generated. Nodes are the entities that exist in a graph, such as a person or a place. Edges represent their relationship – how they are connected to each other. This type of structure allows for more insights to be revealed, the larger and more complex the data. These characteristics are essential in presenting data in a way which makes it easier for AI agents complete tasks in an efficient and reliable way.
Users are finding that GraphRAG’s answers are not only more accurate, but also more complete, faster, and more accurate. GraphRAG can be used by an AI agent to answer customer service questions. The agent will then be able to offer a discounted broadband package based upon a complete understanding about the customer. How long has the client been with the company for? What services do they use? Has anyone ever complained?
In order to answer these questions, nodes are created to represent the customer’s experience with the company, including previous interactions, service usage and location, and edges to show them the cheapest or the best service. A fragmented view of the data may lead to an agent offering a discounted package that was not due. This could have cost implications for your business.
According to the CEO of Klarna: “Feeding a LLM with fragmented and dispersed corporate data will lead to a very confusing LLM”. The outcome is different when data is connected into a graph. Positive results have been reported, such as LinkedIn’s Customer Service team who have reduced median resolution time per issue by 28.6% after implementing GraphRAG.
Why connected data is critical to Agentic AI readiness.
The LLMs that power AI agents are improving rapidly with each iteration. Agentic frameworks make it easier to create complex, multi-step apps. The next step is to make enterprise data rich, connected and contextually aware so that it can be fully accessible to these powerful AI agents. This step will allow businesses to unlock their data’s full value, enabling agents who are not only more efficient and accurate but also easier to explain and understand. The integration of Agentic AI with knowledge graphs is transformative in this case. Data that is connected gives agents the context to think more clearly and produce smarter outputs.
Here’s a list of some of the best survey tools.
The article was produced for TechRadarPro’s Expert Insights Channel, where we feature some of the brightest minds working in the technology industry. The views expressed are those of the writer and not necessarily those TechRadarPro, Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro
