AI technology continues to develop at an extremely fast pace, and the next AI keyword after ChatGPT and Large Language Modeling (LLM) is “AI Agent,” which can independently accomplish complex tasks.
What is an AI agent? How is it different from past AI tools?
If we use the human brain as a metaphor, the book “Thinking, Fast and Slow” divides human thinking into two systems: System 1 relies on intuition, is good at making associations, and is able to react to the situation at hand, but is prone to overgeneralization, which may lead us to make wrong decisions; while System 2 is more planned and logical, and can determine what to do after careful examination, but in a slower and more labor-intensive manner.
In the past, AI generation tools led by ChatGPT were closer to the intuitive thinking of System 1, which often led to “hallucinations” and limited the application of AI tools in real life. In contrast, AI agents are closer to System 2 and can autonomously complete the analysis, memory, planning, and execution steps needed to solve problems without any human intervention.
Take personal secretaries as an example. In the past, generative AI tools could only plan itineraries, but AI agents can complete follow-up actions such as booking air tickets, booking hotels, and getting tickets for attractions on the itinerary. Taking sales development representatives as an example, an AI assistant can only help draft development letters to prospective customers, while an AI agent can independently list potential customers, customize messages for each customer, automatically send letters, and automatically book meeting times based on responses to letters, etc.
Sierra is one of the leading innovators in the field of AI agents. It was founded by Bret Taylor, former CEO of Salesforce and chairman of OpenAI. The AI customer service Sierra has created can use natural-language technology to talk to customers, understand their company’s internal business and operations, and directly solve customers’ needs, such as assisting with returns, issuing receipts, and so on. It has already been introduced by many enterprises.
Let’s try to figure out how much potential AI agents can unlock. According to a 2020 MIT study, a robot in a factory is more than three times as productive as a human employee; as long as there isn’t too much of a difference between the two, just the fact that it can work 24 hours a day without leaving work makes AI more productive.
It is conceivable that in the future, when AI agent technology becomes more mature, companies’ operational efficiency will be greatly improved again. I also believe that the emergence of AI agents will likely change the landscape of future start-ups. When there are AI agents for marketing, business, design, customer service, and other functions that can be imported, a “one-person unicorn” with a very small team, or even a billion-dollar valuation by the founder alone, may soon emerge.
From fast thinking to slow thinking, AI is evolving at an unimaginable pace. However, each ability is not absolutely good or bad but has its own unique advantages and suitability for use in different scenarios. For example, jobs that are routine or require creative thinking may be better suited for fast-thinking AIs with strong reaction and association skills, while slow-thinking AIs with logical thinking skills are needed for planning annual strategies. Those who can utilize these tools in the most powerful combination will be the ultimate winners!
