Last year, Salesforce and Hubspot announced that they were all AI agents. The CEO of Salesforce, Marc Benioff, said that the company was making a “hard pivot” and would be 100% on its agentforce platform.
“”[Agentforce] is the third wave of AI-advancing beyond co-pilotes to a new era of very precise and low hallucination intelligent agents that actively stimulate customer success, “he told DreamForce last September.
So what are the agents?
Here is what marketing media say: agents are an AI that use tools to achieve objectives. These are intelligent systems that perform tasks without human intervention. The agents are characterized by their ability to operate independently, to make decisions and to act.
“Agents are nothing new,” said Christopher Penn, co-founder of Trustinsight and the main scientist of the data.
“”[Agents] are not really agents, ”said Paul Roetzer, CEO of AI Marketing Institute.
Dig more deeply: Watch an extract from Martech’s conversation with Christopher Penn
Autonomous applications
So what are they?
Penn considers them as autonomous applications. “They have a linguistic model (LLM) integrated into part of them, as well as infrastructure around them which allow them to operate according to a certain schedule or to have more than one model of tandem”, A -Ali said.
Dig more deeply: What is the difference between agentics and generative AI?
Roetzer thinks it gives them too much credit.
“They just changed the names of the bots,” said Roetzer. The agents were supposed to be much more impressive than what we have now.
“We have been promised something that perfectly uses your computer, by filling tasks such as filling the forms and travel booking with access to your accounts,” he said.
For Roetzer, the current reality is that although they use LLM in their processes, they do not have the complete capacities that would differentiate them from robots.
Let us call this an agent
“Because there is an LLM used in the process, because the AI is used in the steps, I have the impression that everyone has become agree to call it an AI agent” , he said.
Penn essentially agrees with this. He said that agents are applications with an LLM and an integrated infrastructure that allows them to execute a specific calendar, then put the product back to another application that does something with it. The addition of the LLM is an interesting addition to an existing technology. However, “It’s not new. Zapier as a business has essentially had agents “for about a decade.
Dig more deeply: Salesforce Agentforce: what you need to know
“Autogen de Microsoft is an agent executive that allows you to have a model to write something,” he said, “then the second model, in this controlled sequence, reread it and says that you do not ‘ did not follow the assignment. ”
These warnings do not mean that these “agents” are not incredibly useful. As Penn points out, these things are powerful and simple in what you can do with it. As a result, they apparently have
An infinite number of use cases. »»
“I suspect that a very expensive consultant somewhere wanted to increase his billing rate and found agent applications, as opposed, applications,” he said.
How to recognize a real agent
You know a real agent because he will not need human interaction.
“But, in the foreseeable future, man will probably be strongly involved in most agent requests,” said Roetzer. “It seems that this year will be many people who make a lot of constructions of tailor -made agents which are essentially a bunch of statements if / then crushed with LLM.”
To be clear, real autonomous agents arrive. Improvement of reasoning capacities will eventually go through a chain of thoughts. This, and the improved multimodal models that can produce text, images, video and audio means that agents will ultimately be more than what has been promised to the origin.
