No one knows what an AI agent is
Silicon Valley is optimistic about AI agents. OPENAI CEO Sam Altman said agents “Will join the workforce” this year. Microsoft CEO Satya Nadella predicted that agents will replace certain knowledge work. The CEO of Salesforce, Marc Benioff, said that Salesforce’s objective was to be “the number one supplier of digital workforce in the world“Via the various” agent “services of the company.
But no one may seem to agree on what an AI agent EastExactly.
In recent years, the technological industry has boldly proclaimed that the “agents” of the AI - the last word in fashion – will change everything. In the same way as chatbots have like OPIC’s AAIP We have given new ways to surface information, the agents will fundamentally change the way we approach work, claim CEOs like Altman and Nadella.
It may be true. But it also depends on how we define “agents”, which is not an easy task. Like the other Jargons linked to AI (for example, “Multimodal”, “AG” and “AI” itself), the terms “agent” and “agency” are diluted to the point of non-insignificant.
This threatens to leave Openai, Microsoft, Salesforce, Amazon, Google and the countless other companies that build gigantic whole products around agents in an annoying place. An Amazon agent is not the same as an agent of Google or any other supplier, and this leads to the confusion – and the frustration of customers.
Ryan Salva, Product Product Director at Google and a former co -pilot chief, said that he had just “hated” the word “agents”.
“I think our industry is holding the term” agent “to the point where it is almost absurd,” said Salva to Techcrunch in an interview. “”[It is] One of my black animals.
The agent’s definition dilemma is not new. In a play last year, the former journalist of Techcrunch Ron Miller request: What is an AI agent? The problem he has identified is that almost all business construction agents are approaching technology differently.
This is a problem that has recently worsened.
This week, Openai published a blog article This has defined agents as “automated systems that can accomplish the tasks in the name of users”. However, in the same week, the company released Developer documentation that agents define as “LLMS equipped with instructions and tools”.
Leher Pathak, Lead de Marketing de Product Api d’Openai, said later in a Publish that it included the terms “assistant” and “agents” as interchangeable – blurring water more.
In the meantime, Microsoft blogs try distinguish Between agents and AI assistants. The first, which Microsoft calls “new applications” for a “world powered by AI”, can be adapted to particular expertise, while assistants simply help general tasks, such as the editorial staff.
AI Lab Anthropic addresses the Méli-Mélo of the definitions of agent a little more directly. In a blogAnthropic says that agents “can be defined in several ways”, including both “fully autonomous systems that operate independently over prolonged periods” and “normative implementations which follow predefined workflows”.
Salesforce has what is perhaps the widest definition of the “agent” of the AI. According to the software giant, The agents are “A type of […] System that can understand and respond to customer requests without human intervention. “The company’s website lists six different categories, ranging from” simple reflex agents “to” agents based on public services “.
So why chaos?
Well, agents – like AI – are a nebulous thing, and they are constantly evolving. OPENAI, Google and Perplexity have just started to send what they consider their first agents – Openai operator,, Google Project MarinerAnd Perplexity shopping officer – And their capacities are everywhere on the map.
Rich Villars, a global research GVP in IDC, has noted that technological companies “have a long history” not to respect the technical definitions.
“They care more about what they are trying to accomplish” at the technical level, said Villars in Techcrunch, “in particular in rapid development markets.”
But marketing is also to be largely blamed, according to Andrew NG, the founder of the learning platform for AI Deeplearning.ai.
“The concepts of” agents “and workflows” agent “had a technical meaning,” said NG in a recent interview, “but about a year ago, marketing specialists and some large companies obtained a hand.”
The absence of a unified definition for agents is both an opportunity and a challenge, says Jim Rowan, head of AI for Deloitte. On the one hand, ambiguity allows flexibility, allowing companies to personalize the agents to their needs. On the other hand, this can – and no doubt already – lead to “poorly aligned expectations” and difficulties in measuring the value and return on investment of agent projects.
“Without a standardized definition, at least within an organization, it becomes difficult to compare performance and ensure coherent results,” said Rowan. “This can lead to various interpretations of what AI agents should provide, potentially complicating the objectives and results of the project. In the end, although flexibility can stimulate creative solutions, a more standardized understanding would help companies better navigate the landscape of the AI agent and maximize their investments. »»
Unfortunately, if the dismantling of the term “AI” is an indication, it seems unlikely that the industry will merge around a definition of “the agent” as soon as – if ever.
