AI Agents: Unveiling the Mystery

# agent

The tech world is buzzing with excitement over AI agents. Big names like OpenAI's Sam Altman, Microsoft's Satya Nadella, and Salesforce's Marc Benioff are all in on it. Altman says agents will "join the workforce" this year, while Nadella thinks they'll take over some knowledge work. Benioff wants Salesforce to be the top dog in digital labor through their "agentic" services. But here's the catch: nobody can quite pin down what an AI agent actually is. Over the past few years, tech companies have been hyping up AI "agents" as the next big thing. Just like how AI chatbots like ChatGPT changed how we get information, these agents are supposed to shake up the way we work, according to CEOs like Altman and Nadella. Sounds promising, right? Well, it all hinges on what you think an "agent" is, and that's not easy to nail down. Terms like "multimodal," "AGI," and "AI" itself are already pretty vague, and now "agent" and "agentic" are joining the club. This vagueness is a headache for companies like OpenAI, Microsoft, Salesforce, Amazon, and Google, who are betting big on agents. Each company's version of an agent is different, which just confuses customers. Ryan Salva, a senior director at Google and former GitHub Copilot leader, is fed up with the term "agents." He told TechCrunch, "I think that our industry overuses the term 'agent' to the point where it is almost nonsensical. It's one of my pet peeves." This confusion isn't new. Last year, former TechCrunch reporter Ron Miller asked, "What's an AI agent?" The issue is that every company building agents does it differently. Things have only gotten messier lately. This week, OpenAI described agents as "automated systems that can independently accomplish tasks on behalf of users" in a blog post. But then, in their developer documentation, they called agents "LLMs equipped with instructions and tools." Leher Pathak, OpenAI's API product marketing lead, added to the confusion by saying on X that "assistants" and "agents" are basically the same thing. Meanwhile, Microsoft tries to draw a line between agents and AI assistants. They see agents as specialized "new apps" for an "AI-powered world," while assistants just help with general tasks like drafting emails. Anthropic, another AI lab, admits that agents can be defined in many ways, from fully autonomous systems to those following predefined workflows. Salesforce goes even broader, calling agents "a type of system that can understand and respond to customer inquiries without human intervention." They list six different types, from "simple reflex agents" to "utility-based agents." So, why all the confusion? Well, agents, like AI, are hard to pin down and always changing. OpenAI, Google, and Perplexity have just started rolling out what they call their first agents—OpenAI's Operator, Google's Project Mariner, and Perplexity's shopping agent—and they're all over the place in terms of what they can do. Rich Villars from IDC points out that tech companies often don't stick to strict definitions. "They care more about what they are trying to accomplish," he told TechCrunch, especially in fast-moving markets. Marketing is also a big part of the problem, according to Andrew Ng from DeepLearning.ai. He says, "The concepts of AI 'agents' and 'agentic' workflows used to have a technical meaning, but about a year ago, marketers and a few big companies got a hold of them." Jim Rowan from Deloitte sees the lack of a clear definition as both a chance and a challenge. It lets companies tailor agents to their needs, but it can also lead to "misaligned expectations" and make it hard to measure the value of agent 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. Ultimately, while the flexibility can drive creative solutions, a more standardized understanding would help enterprises better navigate the AI agent landscape and maximize their investments." Sadly, if the way the term "AI" has evolved is any guide, it looks like we won't see a single definition of "agent" anytime soon—or maybe ever.

Related article

Comments (20)