Agentic AI: Why Software Integration is Essential to Unlocking the Next AI Frontier
To say that technology is evolving fast is an understatement. No sooner have we started to grasp the potential and pitfalls of Generative AI in the legal space than the next development looms on the horizon. Agentic AI represents the evolution of workplace AI, ushering in an environment where AI doesn’t simply assist with tasks but instead completes them autonomously and independently of human input. It promises to deliver enormous productivity, efficiency, and accuracy gains in data-intensive industries ranging from healthcare and autonomous vehicles to finance and customer service – and there seems little doubt that the legal sector will be among them.
This blog takes a look at Agentic AI, how it could be applied in the IP management space and what needs to be in place before this can happen.
Agentic AI vs Generative AI: From Making to Doing
First, it is worth articulating the difference between Generative AI and Agentic AI. At the most basic level, the clue is in the name: Generative AI makes things, Agentic AI does things. But, of course, there is more to it than that.
Generative AI creates new content, such as text, images, music, or even code, based on the data it has been trained on. Unlike ‘traditional AI’, which focuses on recognising patterns or making predictions, generative AI creates outputs that mimic the characteristics of the training data while introducing new and unique elements.
However, legal documents are not always the best places to introduce “new and unique” elements. Several incidents of generative AI hallucination in the form of non-existent case law citations have hit the headlines, which is unacceptable in an industry that relies on accuracy. Nevertheless, despite this inauspicious start, Generative AI will find its place in law firms and in-house teams.
Agentic AI is a different category. It is a system composed of small, specialised pieces of software that act together or independently to achieve system objectives, drawing on the reasoning and discovery capabilities of large language models (LLMs) to determine the best approach. Agentic AI is autonomous and capable of making decisions, learning, and improving over time.
There’s a good primer on Agentic AI on the UK government website. It illustrates its capabilities using the example of an online ordering system where the Agentic AI system is given the objective of “fulfilling the order”. Traditional process-based systems can create the order, add items, initiate the order picking system, and dispatch items. An Agentic AI system can do this, but it can also take more dynamic actions, such as checking the latest pricing and deciding to offer a discount to build customer loyalty, for example. It can react to events that occur during the order process, such as an item going out of stock, by initiating a back-order and communicating this to the customer. Over time, the agent can build a profile of the customer, enabling it to offer a service that reflects the customer’s preferences.
In contrast to Generative AI, Agentic AI is more akin to a very advanced form of process automation augmented with the ability to act on context derived from changing data and circumstances and make decisions about the best way forward. Thus, when examining its applications, processes are a good place to start. Could Agentic AI handle trade mark management?
Agentic AI for trade mark management
The trade mark lifecycle has many stages and processes, from creative concept to screening and clearance searches, through formal applications, registration, maintenance, and ongoing monitoring and protection.
Different forms of Artificial Intelligence could be employed at each stage to facilitate the process, and it is possible to imagine a future where Agentic AI – with its ability to analyse data and act on the results – could ultimately make much of the process autonomous and iterative – albeit with the essential human-in-the-loop providing critical oversight. In the same way that the order system example can react to events along the way, making decisions to assist it in achieving its goal of delivering goods to the customer, the intelligent trade mark agent could handle bumps on the road to achieving and maintaining a registered mark.
This won’t happen in the immediate future – although IBM is already marketing AI agents for arguably simpler departments such as HR and Sales – but it is an interesting concept, and it does start us thinking about the foundations needed before autonomous activity on this scale could take place.
Preparing the IPTech ecosystem: software integration and data accuracy
For an AI agent to act successfully, it will need to be able to access the different tools and databases that comprise the law firm or in-house team’s IPTech ecosystem. These could include IP Management Systems like CedarIP, search and watch tools like Haloo or Clarivate, patent drafting tools like Qatent, global historic IP case data and analytics like Darts IP, renewal service providers such as Patentrenewal.com and SunIP, and online brand protection solutions like Corsearch. For that to happen successfully, these tools will need to be accessible to the agent via Application Programming Interfaces (APIs) and built on a technology foundation that supports AI integration and interrogation.
The same will need to be true of the external databases that the agent accesses as it pursues its objective. These include national trade mark registries and international systems such as the WIPO and EU IPO. Several national IP offices are working to introduce AI features – mostly traditional AI at this stage – to assist employees by increasing efficiency and accuracy. The UK IPO, for example, is undergoing a comprehensive One IPO transformation programme that will, among other things, ensure that APIs are exposed so external IP software can interact directly with its systems.
AI-powered autonomous decision-making will only deliver satisfactory outcomes if the data it is working from is accurate and accessible. As the saying goes: garbage in, garbage out – and this applies at scale to AI. Many organisations will need to undertake a considerable data identification and cleansing activity in order to fully leverage it. Those with an established and well-maintained IP management system will have an advantage, as their trade mark records should be well-ordered and logically stored. However, there may be other databases that an AI agent needs to access as it seeks to accomplish its task, and these must also be accurate and up-to-date.
Conclusion: “to be continued!”
In summary, while Agentic AI offers exciting potential for automating complex processes like trade mark management, its success hinges on robust integration with existing tools, accurate and accessible data, and a well-prepared IPTech ecosystem. Although full autonomy may be a distant goal, laying the groundwork today through improved APIs, data management, and system compatibility will pave the way for this transformative technology. There’s no doubt the story of Agentic AI in IP management is to be continued!
Further reading:
https://legal.thomsonreuters.com/blog/the-key-to-autonomous-legal-workflows-with-agentic-ai/