The buzzwords that have grown to the point of meaninglessness are as old as the tech industry itself. The top one is something like an “AI Agent” and its variations (agents).
So of course, no one really knows what an AI agent is. Even people with software engineering backgrounds working at Andreessen Horowitz say there is no agreed definition, even Andreessen Horowitz, who is crazy funding AI startups.
Three A16Z Infrastructure Investment Partners – Guido Appenzeller, Matt Bornstein and Yoko Li-, “What is an AI Agent?”
For perspective, A16Z, a backer of hot AI companies such as Openai and Anysphere (the maker of Cursor), is reportedly a very ganho about AI opportunities and is trying to raise $2 billion megafunds to invest even more in the sector. In September, two other A16Z VCSs explained the company's excitement, writing on their corporate blog:
The AI startup “continuum” describes their products as agents to win buzz, says Appenzeller.
“The easiest thing I've heard of being called an agent is basically some kind of knowledge base plus a smart prompt,” he said. This so-called agent receives questions from humans and then gets a “canned” response.
However, these days, companies that want to manufacture agents or create them have described them as alternatives to human workers.
TechCrunch Events
Berkeley, California | June 5th
Book now
To do that, Appenzeller says that AI software must be “close to AGI.” That is, “it has to last for a long time” and “it has to work independently on the issue.”
But such things “it's not working yet,” both he and Li said.
In reality, ensuring that this early AI agent technology works well has been an incredibly difficult journey, Jaspar Carmichael-Jack, CEO of Artisan, a Sales AI agent company, told TechCrunch last month. Carmichael Jack is still hiring humans despite the startup's viral “stop hiring humans” ad campaign.
For AI to become a replacement for true human workers, there are important technical issues to be resolved, such as persistent long-term memory (and associated costs) and hallucinations. This is because no company wants to hire employees (human or artificial) who can't remember previous conversations and lie randomly.
During the podcast, the A16Z trio landed on a solid definition of what is possible today. As Li explained, AI agents are multi-step LLM inference with dynamic decision trees.
In other words, she said agents aren't just bots who do tasks when asked. You must also be able to get a list of prospects from the database, decide which ones to send emails, write emails, and be able to make decisions about tasks and take action autonomously. Alternatively, write your code to determine where to insert it.
As for whether agents can actually replace humans in the near future, I agreed that all three VCs could be used to handle some of the tasks humans do now, as automation always does. However, this can actually lead to companies hiring fewer and fewer human workers as productivity increases.
Bornstein said that considering the time when humans are no longer needed, it is impossible to imagine the times. From our “perch in Silicon Valley,” the tech industry can “forget” that most people have jobs that require human creativity and “thinking.” To replace humans with bots, he said, “I don't know if there is a theoretical possibility.”
Still, this kind of human exchange rhetoric is often done for marketing/business models and pricing reasons, but “it's a big reason for the confusion we're experiencing right now,” says Bornstein.
The result is that if those who see all of the most cutting edge use of AI agents are skeptical of the boldest claims that AI agents make today, that is probably a good indication that the rest of us should be.