The concept of AI agents is still relatively new, and there is no consensus on what exactly constitutes an AI agent. However, at its core, an AI agent can be described as AI-powered software that performs tasks on behalf of humans, automating processes that were traditionally handled by customer service agents, HR personnel, or IT help desk employees. The tasks performed by AI agents can extend beyond simply answering questions and may involve complex problem-solving across multiple systems.
Different companies and experts have varying definitions and use cases for AI agents. Google sees them as task-based assistants, helping with coding, marketing, or IT troubleshooting. Asana views AI agents as extra employees that handle assigned tasks. Sierra, a startup founded by industry veterans, sees agents as customer experience tools that solve complex problems.
Despite the lack of a cohesive definition, AI agents share the common goal of completing tasks in an automated manner with minimal human interaction. These systems incorporate various AI technologies such as natural language processing, machine learning, and computer vision to operate autonomously or alongside human users.
Experts believe that as AI technology advances, AI agents will be capable of doing much more on behalf of humans. Factors such as GPU price/performance, model efficiency, model quality and intelligence, and improvements in AI frameworks and infrastructure will contribute to the growth and capabilities of AI agents.
However, it’s important to note that AI faces unique challenges compared to other technologies. MIT robotics pioneer Rodney Brooks highlights the difficulty of crossing systems and the lack of access to legacy systems’ basic APIs. While improvements are being made, there may be limitations to what AI agents can currently achieve.
David Cushman of HFS Research sees the current generation of bots as assistants that help humans complete tasks in pursuit of strategic goals. The challenge lies in enabling machines to handle contingencies in a fully automated way, which is still a work in progress.
The development of an AI agent infrastructure, a tech stack specifically designed for creating agents, is crucial for the proliferation of AI agents. The industry needs to build an infrastructure that supports AI agents and the applications that rely on them. Over time, reasoning and frontier models will improve, and developers will require a platform that offers scalability, performance, and reliability.
It’s worth noting that multiple models, rather than a single large language model, may be necessary to make AI agents work effectively. Fred Havemeyer of Macquarie US Equity Research suggests that the most effective agents will likely be a collection of different models with a routing layer that delegates tasks to the most suitable agent and model.
While the industry is working towards the goal of autonomous agents, there is still a transition period and breakthroughs are needed for AI agents to operate as envisioned. The current advancements are promising, but there is still progress to be made before AI agents can operate independently and handle abstract goals with complete autonomy.
