Gregorio Patiño Zabala, Co-Founder & Head of Business Development – FSI unit at Pragma.
In the mid-1980s, the hype around artificial intelligence entered what we know today as the second AI winter. This new plateau resulted from existing AI systems not meeting expectations, pessimism drying up funding and the internet boom attracting more attention and investment. However, the idea of intelligent agents was born during this time, a paradigm that has recently regained relevance and is already reshaping all kinds of industries.
Back then, intelligent agents were defined as autonomous systems capable of achieving goals without significant human intervention. Instead of being programmed with rigid rules, intelligent agents gather information from their environment and choose the most efficient path to reach their goals. With recent developments in generative AI, the concept of intelligent agents has turned into a tangible reality.
What Kinds Of AI Agents Are Used Today?
According to McKinsey, use cases of generative artificial intelligence, such as AI agents, can add value in several industries. Their contribution to task automation, their capacity to improve customer service, boost productivity and enable focus on more value-added activities are just a few examples of their potential.
A common misconception about intelligent agents is that they are fancy chatbots. The truth is that they go much further. For instance, Salesforce recently announced it will release two new intelligent agents based on its generative artificial intelligence, Einstein.
On the one hand, there is Einstein Sales Coach Agent, a tool that can be trained with company data to simulate a conversation with a customer and allow salespeople to practice and improve their skills. On the other hand, there is the Einstein Sales Development Rep (SDR), which automates much of the interaction with prospects-answering product questions, dealing with objections or scheduling a meeting that will allow sales teams to make more intelligent use of their time.
What do these intelligent agents have in common? At their core is a foundational model like GPT, BERT or Claude, but they go one step further. By using harmonized company data, intelligent agents can have their own knowledge base that, added to their specific purpose, allows them to function as customized and objective-driven versions of the generative AIs on the market.
Additionally, Salesforce and other technology providers have a low-click/no-code approach that enables companies to leverage AI tools without major investments in development. Still, they are not alone in democratizing the power of generative AI.
Amazon Web Services is another company positioning its offering in this arena. AWS Bedrock is a service through which companies can create their own intelligent agents. Thanks to this, organizations can create tailored solutions capable of maximizing their information and technological resources while keeping their data secure.
One of the most common concerns that comes with training intelligent agents is exposing sensitive information to security risks. In response to this, AWS Bedrock has implemented a firm security policy in which data never leaves the client’s control and is never used to train third-party systems. Additionally, it uses tools to transfer information securely and monitor compliance with various regulatory standards.
Another exciting alternative in the world of intelligent agents is OpenAI Azure, which allows using code to “teach” skills to an intelligent agent, such as offering personalized recommendations, answering natural language queries using database information and providing responses based on real-time information. Azure AI services also support multi-agent architectures, where different agents and APIs can be connected to complete more sophisticated tasks.
How To Get The Most Out Of Intelligent Agents
It is true that generative AI cannot be taken as a source of truth. Data security and hallucinations still represent a significant risk for any company that uses AI in its processes.
It has also been rightly pointed out how little impact artificial intelligence has had on the economy’s big picture. However, in my company’s experience creating AI solutions for organizations with large amounts of data—such as those in the financial, insurance and retail industries—we have seen how intelligent agents can boost process automation and productivity.
I’ve recently seen firsthand how companies in these sectors are building intelligent agents. Based on this experience, there are a few recommendations I have for organizations looking to leverage this technology:
• The information with which your agent will be trained must be carefully curated. Irrelevant and outdated information could cause hallucinations and undesirable behavior.
• Intelligent agents use foundational models as a basis. There is a wide variety of these models, and choosing the right one is crucial. More up-to-date versions are usually expensive, but sometimes, earlier versions will provide the required accuracy at a better cost. Also, the choice of model augmentation between GAR, fine-tuning and other techniques will make a difference.
• Identifying highly repetitive tasks and processes based on document and information processing is a great way to focus your automation efforts.
Indeed, we still need to reach a stage where intelligent agents serve any business, no matter where they are in their journey to data maturity. Yet, the autonomy of intelligent agents and the fact that they can be trained securely with in-house information makes them particularly relevant for companies that have already made some progress in data governance and architecture.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
