The tech industry, particularly Silicon Valley, is enthusiastic about the potential of AI agents. According to OpenAI CEO Sam Altman, AI agents will enter the workforce this year. Similarly, Microsoft CEO Satya Nadella predicts that agents will replace certain knowledge work, while Salesforce CEO Marc Benioff aims for his company to be the top provider of digital labor worldwide through its various “agentic” services.

However, there is a significant lack of consensus on what exactly constitutes an AI agent.

In recent years, the tech industry has been abuzz with the concept of AI “agents” as a game-changer. Similar to how AI chatbots like OpenAI’s ChatGPT revolutionized information retrieval, agents are expected to fundamentally alter the way we work, as claimed by CEOs like Altman and Nadella.

The validity of this claim depends on how one defines “agents,” which is a challenging task. The terms “agent” and “agentic” are becoming increasingly diluted, much like other AI-related jargon such as “multimodal,” “AGI,” and “AI” itself, to the point of losing their meaning.

This lack of clarity puts companies like OpenAI, Microsoft, Salesforce, Amazon, and Google, which are investing heavily in agent-based product lineups, in an awkward position. The fact that an agent from one vendor is not the same as another is causing confusion and frustration among customers.

Ryan Salva, senior director of product at Google and former GitHub Copilot leader, expressed his frustration with the term “agents,” stating that the industry overuses it to the point of rendering it meaningless.

“I think our industry overuses the term ‘agent’ to the point where it is almost nonsensical,” Salva said in an interview with TechCrunch. “[It is] one of my pet peeves.”

The issue of defining AI agents is not new. Former TechCrunch reporter Ron Miller raised this question last year, highlighting that nearly every company building agents approaches the technology differently.

The problem has worsened recently, with companies like OpenAI publishing conflicting definitions of agents. In a blog post, OpenAI defined agents as “automated systems that can independently accomplish tasks on behalf of users.” However, the company’s developer documentation defined agents as “LLMs equipped with instructions and tools.”

Leher Pathak, OpenAI’s API product marketing lead, further complicated the issue by stating that she considers the terms “assistants” and “agents” to be interchangeable.

Meanwhile, Microsoft’s blogs attempt to distinguish between agents and AI assistants, with the former being tailored to have specific expertise and the latter helping with general tasks like drafting emails.

AI lab Anthropic acknowledges the complexity of defining agents, stating that they “can be defined in several ways,” including fully autonomous systems and prescriptive implementations that follow predefined workflows.

Salesforce has a broad definition of AI “agent,” describing it as a system that can understand and respond to customer inquiries without human intervention, with six different categories ranging from “simple reflex agents” to “utility-based agents.”

The reason for this chaos is the nebulous nature of agents and their constant evolution. Companies like OpenAI, Google, and Perplexity have just begun shipping their first agents, with varying capabilities.

Rich Villars, GVP of worldwide research at IDC, noted that tech companies often do not rigidly adhere to technical definitions, prioritizing their technical goals instead.

“They care more about what they are trying to accomplish” on a technical level, Villars said, “especially in fast-evolving markets.”

However, marketing also plays a significant role in the confusion, according to Andrew Ng, founder of AI learning platform DeepLearning.ai.

“The concepts of AI ‘agents’ and ‘agentic’ workflows used to have a technical meaning,” Ng said, “but about a year ago, marketers and a few big companies got a hold of them.”

The lack of a unified definition for agents presents both an opportunity and a challenge, according to Jim Rowan, head of AI for Deloitte. While the ambiguity allows for flexibility and customization, it may lead to “misaligned expectations” and difficulties in measuring the value and ROI of agentic projects.

“Without a standardized definition, at least within an organization, it becomes challenging to benchmark performance and ensure consistent outcomes,” Rowan said. “This can result in varied interpretations of what AI agents should deliver, potentially complicating project goals and results.”

Unfortunately, given the industry’s history with the term “AI,” it is unlikely that a unified definition of “agent” will be established anytime soon.