AI agents are being hailed as the next big leap in artificial intelligence, yet there’s no clear consensus on what exactly constitutes an AI agent. At its core, an AI agent can be described as software powered by artificial intelligence, designed to perform tasks traditionally handled by human roles like customer service agents, HR personnel, or IT help desk employees. These agents go beyond answering questions; they can perform complex tasks across multiple systems.
While the concept seems straightforward, the lack of a unified definition complicates the picture. Even among tech giants, there is no agreement. Google views AI agents as task-based assistants tailored to specific roles: coding support for developers, creating color schemes for marketers, or assisting IT professionals in tracking down issues via log data.
Asana sees AI agents as virtual coworkers, efficiently handling assigned tasks. Meanwhile, Sierra, a startup founded by former Salesforce co-CEO Bret Taylor and Google veteran Clay Bavor, positions AI agents as advanced customer experience tools, solving complex problems beyond the capabilities of traditional chatbots.
Despite the varying definitions, the goal remains the same: to automate tasks with minimal human intervention. Rudina Seseri, founder and managing partner at Glasswing Ventures, acknowledges the early stage of AI agents and the resulting ambiguity. “There is no single definition of what an ‘AI agent’ is. However, the most frequent view is that an agent is an intelligent software system designed to perceive its environment, reason about it, make decisions, and take actions to achieve specific objectives autonomously,” she explains.
Seseri highlights that these agents leverage various AI technologies, including natural language processing, machine learning, and computer vision, to operate autonomously or in collaboration with humans and other agents.
Aaron Levie, co-founder and CEO of Box, is optimistic about the future capabilities of AI agents. He believes that advancements in GPU price/performance, model efficiency, quality, intelligence, and AI frameworks will significantly enhance what AI agents can achieve. However, MIT robotics pioneer Rodney Brooks cautions that AI faces more complex challenges than other technologies, and progress might not follow the rapid pace of Moore’s law.
The difficulty of integrating systems, particularly legacy ones without basic API access, presents a significant hurdle. While progress is being made, the ability of software to seamlessly access multiple systems and solve problems in real-time remains a challenge.
David Cushman, a research leader at HFS Research, aligns with Asana’s perspective, viewing current AI agents as assistants helping humans achieve strategic goals. However, he notes the challenge of true automation, where machines handle contingencies independently.
Jon Turow, a partner at Madrona Ventures, emphasizes the need for an AI agent infrastructure, a tech stack designed specifically for creating and supporting these agents. “Our industry has work to do to build infrastructure that supports AI agents and the applications that rely upon them,” Turow writes. He envisions a future where reasoning capabilities improve, frontier models steer workflows, and developers can focus on product and data, relying on a robust underlying platform.
Fred Havemeyer, head of U.S. AI and software research at Macquarie US Equity Research, believes that effective AI agents will likely involve multiple models with a routing layer to delegate tasks to the most suitable agent and model. He sees this as a step toward truly autonomous agents capable of independently achieving abstract goals.
Ultimately, while the industry is moving toward fully autonomous AI agents, we are still in a transitional phase. Significant advancements and breakthroughs are needed to realize the vision of AI agents operating independently as envisioned today. The journey is promising, but it’s important to recognize that we haven’t reached the final destination yet.
