What is an AI agent? AI agents are everywhere, with some news outlets proclaiming them to be the future of AI. We don’t disagree. At Infinitus, we talk a lot about our AI agent and the technology that powers it. Also known as Eva, the Infinitus AI agent makes phone calls to commercial and government payors on behalf of healthcare providers, performing benefit verifications, and prior authorization and prescription status checks. The calls our AI agent makes are complex, sometimes lasting an hour or more, and its ability to scale often makes Eva feel like the most productive member of a team. But what is an AI agent, exactly? What an AI agent is not: A chatbot or a voicebot Let’s start out with a common misconception: AI agents = chatbots. The reality? They’re not synonymous. While chatbots and AI agents are both generative AI in the sense they generate text or speech, there are a number of key differences in the experience they create and how they are developed: AI agent Voicebot Chatbot Medium Audio or phone calls Audio or phone calls Text or messaging Speed Must understand and respond in milliseconds, or the interaction will fail (one party will hang up) Must understand and respond in milliseconds, or the interaction will fail (Caller repeatedly screams “Agent” or presses “0”) Must understand and respond in seconds or minutes (insert bubbles) Task complexity Freeform conversation: Must accurately predict the next right action (e.g., answer, clarify, redirect) IVRs or Voicebot: Predefined list of options: “What can I help you with today” “Choose from one of the following you can Press # or say [ ]” Predefined list of options: e.g., “What can I help you with today?” “Choose from one of the following” ; Often the “other” option sends user to a “contact us” form. Interaction length Minutes to hours Seconds to minutes Seconds to minutes AI tech most commonly used Natural language processing (NLP) AI solutions, including but not limited to: • Automatic speech recognition (ASR) or Speech to Text (STT) • Text to Speech (TTS) • Multimodal (Audio + Text) • Shallow models • Large language models (LLMs) • Foundational models (eg. OpenAI GPT, Meta Llama) Natural language processing (NLP) AI solutions, including but not limited to: • Decision trees or flow charts of steps • Automatic speech recognition (ASR) or Speech to Text (STT) • Text to Speech (TTS) Natural language processing (NLP) AI solutions, including but not limited to: Decision trees or flow charts of steps, Foundational models (eg. OpenAI GPT, Meta Llama) Who can build Data scientist Engineering Line of business user (eg. Contact center manager) Line of business user (eg. Marketing manager) User expectation Sounds like a human, so therefore it should have a higher level of accuracy Sounds like a robot, and historically is poor quality, so while frustrating it isn’t Acts like a robot, so therefore it isn’t expected to have a higher level of accuracy AI agents are capable of more than just simple question-and-answer interactions. An AI agent like the one Infinitus offers is not constrained to simple responses but can act autonomously, actively completing more complex, nuanced, and/or lengthy tasks. As Wired magazine described it, AI agents don’t “just [provide] answers or advice about a problem presented by a human,” but instead can take action to solve it. Which sets us up nicely for the next important thing to know about AI agents. How does an AI agent work? An AI agent is a program or system that perceives its environment and takes actions to achieve specific goals or objectives. It works by gathering information about that environment, then leverages that information combined with algorithms or models, along with a knowledge base, to determine the appropriate action to take (e.g., answer question, clarify question, clarify answer, redirect to another resource, etc).The most advanced AI agents learn from past actions, and adapt their “behavior” based on learnings. While the specific workings of different AI agents can vary significantly, the general principles of perception, reasoning, action, learning, and adaptation are common to most AI agents. At Infinitus, we employ a purpose-built AI system, which is multi-model and multimodal in nature, and uses different model architectures, training methods, and our knowledge base depending on the AI Agent’s objective at hand. We have a combination of proprietary in-house trained models and open-source models that our team has fine-tuned and/or built on top of. The parameters of our models range from dozens in our more shallow models and 100 million for our larger in-house models, to as much as 10 trillion in the GenAI models that we employ. In addition, many AI agents – especially in areas like healthcare where accuracy is paramount – incorporate a human-in-the-loop element. More on that in a bit. An AI agent can drive a conversation Think about the last time you engaged with a chatbot. Maybe it was ChatGPT or Google’s Gemini, or maybe it was a helpful bot that popped up in a window in the lower right-hand corner of a company’s website. In these examples, it’s likely you were seeking information, and therefore leading the interaction. Similarly, until now, most AI agents responded to inquiries. But at Infinitus, our AI agent is in the driver’s seat of a conversation. Something that sets the Infinitus AI agent apart is its ability to lead rather than follow, especially in uncertain or unpredictable circumstances. The Infinitus AI agent dictates the flow of a conversation, because it is knowledgeable about the information required, and assertive enough to seek it in a structured format. An AI agent may have human ‘guardrails’ Despite the advanced capabilities of AI agents, they are not infallible. Recognizing this, Infinitus employs a system where experienced healthcare professionals “oversee” the AI agent. These experts provide institutional knowledge and assistance when the AI agent encounters an impasse or detects inaccuracies. Such human oversight ensures that the Infinitus AI agent operates at a high level of accuracy, critical because mistakes could lead to serious consequences like delayed patient treatment or financial challenges for providers. In healthcare especially, it’s likely any AI agent would employ humans in the loop. The benefits of a healthcare AI agent There are myriad benefits to working with an AI agent in healthcare. For pharmaceutical companies, improved data accuracy can help boost the efficiency of patient support programs. That improved accuracy also leads to fewer wrongful claims denials, which results in meaningful cost savings. Across the entire healthcare landscape, an AI agent can help increase employee productivity by empowering organizations to accomplish more with fewer resources. The Infinitus AI agent helps our customers save their staff 30 minutes per call they would otherwise spend on the phone – that’s time that can be put to use on empathetic tasks better completed by a human. It improves the accuracy of data collected by 10% when compared to human callers. If you’d like to learn more about the Infinitus AI agent, or see it in action on a real call, watch a benefit verification automation demo. Or, to learn more about what it’s like to work with the Infinitus AI agent, hear directly from Fortrea Patient Access.
