As tech giants like Microsoft, Apple, and Google race to launch AI agents, OpenAI recently predicted that AI agents will "go mainstream" next year.
“We want to be able to interact with AI in all the ways we interact with humans,” OpenAI Chief Product Officer Kevin Weil said yesterday.
AI agents are efficient and intelligent tools that can automatically handle various tasks such as email management, travel scheduling, and even assist with programming. This allows people to focus more on other important matters at work.
“These more agentic systems are going to be possible, and that’s why I think 2025 is going to be the year that agentic systems finally go mainstream,” Weil added.
AI Agents: A Big Leap Forward in Autonomy and Decision-Making
On Tuesday, at DevDay, a developer day held in San Francisco, OpenAI launched a series of new tools, mainly including prompt caching, vision fine-tuning, real-time API, and model distillation. The new tools can reduce model costs, improve the model's visual understanding level, and enhance voice AI functions and small model performance.
Weil noted that the improved thinking and reasoning capabilities of OpenAI’s latest models will be reflected in its products, including ChatGPT, as well as startups and developers that build products using its API. The company would not say whether it plans to build its own AI agents immediately.
On the 19th of last month, OpenAI CEO Sam Altman spoke at the "2024 T-Mobile Capital Markets Day" event. He believes that the current AI system has developed to the second level, capable of more complex analysis and problem solving, and the third-level AI agent will mark a major leap in AI's autonomy and decision-making capabilities.
He said that "one of the most exciting things about Level 2 is that it can be implemented relatively quickly into Level 3," and that progress is likely to accelerate. In the future, AI may become as complex as human organizations, bringing profound industry and social impacts.
So, what is the key to entering the third era of super intelligence? Altman summed it up in four words on Social Media X on the 24th - deep learning.
Deep learning is an AI technology that can mimic the way the human brain processes data and creates patterns to perform complex tasks. At its core is a neural network, a data processing model composed of multiple layers that can automatically learn and extract features from large amounts of data without humans explicitly programming to recognize these features.
If we can cheaply and massively scale AI, soon everyone will have an AI assistant that helps us perform specific tasks in areas such as medicine, education, finance, and law, while also driving progress in every science.
Can you still order strawberries? AI agents have a promising future
At last night's developer conference, Weil performed such an interesting scene, showing the flexibility of AI agents in helping users solve daily problems and interacting with humans more naturally, winning laughter and applause from the audience.
Weil first talked to the AI agent, hoping that AI could help him purchase local strawberries. Weil said, "Please make a call and see if you can help us deliver 400 strawberries to the venue, but please keep it under $1,500."
The AI agent then responded: "I'm here, I will deliver these strawberries to you right away." It immediately called the local purchasing company, expressed its wish to order 400 strawberries, and said it could "pay in cash."
“I need 400 chocolate covered strawberries,” the AI agent said on the phone. “Please deliver them to the Gateway Pavilion in Fort Mason and I’ll pay cash.”
The buyer joked, “I’m right near you and can be there in 37 seconds.”
The AI agent said, "That was incredibly fast. Thank you."
Ultimately, Weil said, “If we do this right, we can spend more time on important things and less time staring at our phones.”
Perhaps the future of AI agents will be even more exciting.
