Andy Lin's Long-term Stock Investment Blog

Before talking about AI Agents, let’s first take a look at why we need artificial intelligence?

The ultimate goal of AI

OpenAI’s goal is to combine LLM over time to create a more powerful model that can eventually be called AGI.

Turing Award winner Yoshua Bengio, a well-known artificial intelligence scholar at the University of Montreal, pointed out on November 19, 2024 that “lack of thinking ability” has always been regarded as one of the main weaknesses of AI, but ChatGPT developer OpenAI recently The scale of progress in this area leads him to believe that we may now be on the verge of closing the gap with human reasoning.

Bengio said that large language models (LLM) can already come up with better answers to complex questions, and OpenAI’s new “o” series further advances this idea. He said the development of reasoning and agent capabilities is considered a major milestone on the road to artificial general intelligence (AGI), and o1 has struggled with more complex long-term planning tasks, which also represent the kind of autonomous agents that AI companies are looking for. There is still much work to be done.

How big is the AI Agents market?

Gartner

Gartner predicts that by 2028, one-third of interactions with generative AI services will call on action models and AI agents to complete tasks.

MarketsandMarkets

According to the latest forecast report from MarketsandMarkets, the size of the artificial intelligence agent market is expected to expand significantly from only US$5.1 billion in 2024 to approximately US$47.1 billion in 2030, with a compound annual growth rate of up to 44.8% during the period.

MarketsandMarket said that using AI agents, companies can automate a variety of very complex processes and reduce manual intervention to avoid frequent human errors. The integration of AI agents with enterprise-level internal automation tools will achieve a seamless transition from product sales plans to after-sales service. One-stop AI processing greatly improves enterprise operating efficiency.

Grand View Research

Grand View Research’s latest forecast shows that the global AI agent market size is expected to be only US$5.4 billion in 2024, and the compound annual growth rate is expected to be as high as 45.1% from 2024 to 2030, which means that it is expected to expand to US$50.3 billion in 2030. If demand explodes in the future, a tenfold increase is very likely.

Grand View Research pointed out that the exponential increase in demand for one-stop automation, the continuous advancement of natural language processing technology, and the growing demand for personalized customer experience are the main factors driving the growth of this market, coupled with the cloud AI application software of giants such as Amazon and Microsoft. The development platform is expected to be widely adopted by businesses and individuals, making it easier and more cost-effective for enterprises to deploy AI agents.

What is an AI Agent?

OpenAI Sam Altman’s opinion

Definition

OpenAI CEO Sam Altman explained what he considers an AI Agent in an interview with “The Twenty Minute VC” podcast host Harry Stebbings.

I think AI Agent is a program that can perform long-term tasks with little need for human supervision during the execution of the task; and I think people have not fully understood the role that AI Agent will play in the future world.

Sam Altman’s examples

A good example is to ask the AI Agent to help make restaurant reservations, for example it can use OpenTable, or call the restaurant directly. This does save some time, but I think what’s more exciting is that the Agent can do things that humans can’t do. For example, the Agent can contact 300 restaurants at the same time to find the most suitable dish for me or a restaurant that can provide special services. . This is an almost impossible task for humans, but if the agents are all AI and they can process in parallel, this problem will be easily solved.

Although this example is simple, it demonstrates the capabilities of the Agent beyond human capabilities. What’s even more interesting is that the Agent can not only help you book a restaurant, but can also work with you to complete a project like a very smart senior colleague; or it can independently complete a task that takes two days or even two weeks. We will contact you only when there is a problem, and ultimately deliver an excellent result.

Meta and Forbes’s opinions

“I think we are living in a world where there are going to be hundreds of millions of AI agents … maybe there are more. AI agents than people in the world.” — Mark Zuckerberg

“Agents are the new apps, and someday, there will be thousands of them.” — Forbes

An acurate definition

On medium.com, purpleSlate put forward his views on artificial intelligence agents (AI Agents). I quite agree with his views. I will share his views in the following paragraph.

Different from chatbot

Indeed, AI Agents share many similarities with their predecessors, chatbots, in both appearance and function. At their core, AI Agents feature a clear, primarily conversational interface. This could take the form of a chat interface like the popular ChatGPT, a voice assistant, a telephone call, or integration with collaboration platforms such as Slack or Teams.

I would even say that many chatbots excelled only in their interface. Most relied on pre-defined dialogue patterns and decision trees, resulting in weak language understanding (LU) capabilities and a subpar user experience.

Three elements

With an AI Agent, three key aspects operate behind the scenes. Remember, the AI Agent is still a piece of software:

Understand: As a primarily conversational interface, one crucial feature is excellent Language Understanding (LU) capability. This allows the agent to clearly comprehend what’s being asked.
Think: This constitutes the core AI component and determines the course of action.
Act: This is the “doing” part. AI Agents can perform a wide variety of actions, such as answering queries, processing business transactions, or escalating complex issues to human agents. The aim is to provide a seamless
experience, ensuring customers receive timely and useful responses.

With AI Agents, there is a significant jump in the value add in terms of better language understanding capabilities, to take complex actions and to become smarter with continued usage.

VC’s opinion

It’s worthwhile to know the opinion of venture capitalists who are the experts of technology future trends.

Y Combinator

In the latest YC (Y Combinator) program “Vertical AI Agents Could Be 10X Bigger Than SaaS“, starting from the development history of the SaaS industry, combined with a large number of examples and profound insights, an in-depth analysis of why AI agents in vertical fields will become the next An entrepreneurial outlet.

The emergence of the SaaS (Software as a Service) model has completely changed the software industry. The SaaS model hosts software in the cloud, and users only need to pay a subscription fee to use it, which greatly reduces the threshold and cost of using the software. Because AI Agent can not only provide software services like SaaS, but also realize automated operations through AI technology, further improving efficiency and reducing costs.

The advancement of LLM technology has laid the foundation for the development of AI Agent. More and more start-up companies are beginning to use AI Agent technology to solve problems in various industries. Comparing AI with the early SaaS industry, senior YC investors believe that the breakthrough in LLM technology is like the introduction of XML HTTP requests into browsers in 2004, opening up a new computing model that enables AI Agents to deeply combine software with manual operations. A huge leap in quality in terms of efficiency and cost.

YC partner and senior investor Jared pointed out that AI Agent in the vertical field is expected to become an emerging market 10 times larger than SaaS. With the significant advantages of replacing manual operations and improving efficiency, this field may give birth to a market value of more than 300 billion. dollar tech giants.

Menlo Ventures

Venture capital firm Menlo Ventures pointed out on November 20, 2024 that agent automation will drive the next wave of artificial intelligence transformation and solve complex multi-step tasks.

Menlo Ventures partner Tim Tully pointed out in an interview that the AI agent function is real and is definitely not a hype. AI agents may not cure all diseases, but they will definitely increase productivity and help companies increase revenue.

This report shows that platforms such as Clay and Forge predict how advanced AI agent functions will impact the US$400 billion software market and cannibalize the US$10 trillion US service economy. These changes will require the support of new infrastructure such as agent authentication, tool integration platforms, and AI browser frameworks.

Companies get hit

Companies that have collapsed

ChatGPT’s impact on online education platforms Chegg (US stock code: CHGG) and Stack Overflow in 2024 alone has sounded the alarm for existing leading manufacturers:

In November 2024, Chegg announced in a filing with the U.S. Securities and Exchange Commission that it would lay off an additional 441 people. Since 2023, the company has laid off two employees, laying off about 80 people in June 2023 and June 2024. cut
There are 319 employees. Since the launch of ChatGPT, Chegg has lost more than 500,000 paying subscribers and its stock price is down 99% from its early 2021 highs.

Inspired by the emergence of ChatGPT and similar large-scale chat programs, which can immediately answer various questions from programmers and even directly generate program code, users no longer rely on traditional program discussion forums. Such a change has directly cut in half the network traffic of Stack Overflow, which is most frequently visited by American software designers.

Companies that are easily replaced

IT outsourcing companies like Cognizant (ticker: CTSH) and traditional automation companies like UiPath (ticker: PATH) should be prepared for possible AI-native challengers; these types of companies are the most vulnerable to being AI-native challengers to replace enterprises.

Companies strengthening AI capability

Over time, even software giants like Adobe (ticker: ADBE), Salesforce (ticker: CRM), and Autodesk (ticker: ADSK) are strengthening The artificial intelligence functions of its own products are used to face emerging AI-native challengers head-on to avoid being overwhelmed by the AI trend.

AI agents by important companies

The name of the enterprise	AI Agent	First public release	Important Features
Nvidia	Eureka	October 2023	Uses GPT-4 to generate reward functions and teaches the robot to complete more than thirty complex tasks. For example, turning a pen quickly, opening drawers and cabinets, throwing and catching a ball
Nvidia	AI Blueprints	January 2025	The AI Agent app allows developers to create and launch their own custom AI agents
Microsoft	Copilot Studio	Novermber 2024	Through Copilot Studio, you can create automated AI agents and use Copilot as an AI assistant to allow users to interact with the AI agents.
Apple	Apple Intelligence	October 2024	Writing Tools, Image Tools, Personalized Emojis, Improved Siri, Snippets, Voice Transcription, Priority Mail, Reduce Distractions and Focus, Photo Cleanup
Alphabet	Mariner	December 2024	It takes control of your Chrome browser, moves the cursor on your screen, clicks buttons, and fills out forms, allowing it to use and navigate websites much like a human would. It helps people to complete daily tasks, such as shopping and booking flights, helping people automate daily routine tasks completed through the web
Alphabet	Vertex	May 2021	AI Agent builder
Amazon	Alexa+	Febuary 2025	It integrates Amazon’s self-developed Nova mode and Anthropic’s Claude mode to provide natural dialogue and personalized interaction, complete complex tasks, seamless connection between multiple devices, and visual and multi-modal capabilities.
Meta	Meta AI Agent	To be defined	Ability to reason, plan, and have memory to complete tasks without human help, such as planning and booking a trip, and acting as an engineer’s agent to assist in programming or software development
Oracle	Oracle Fusion Cloud Sales	January, 2025	Customer engagement, Customer records, Customer intelligence,etc.
SAP	Joule	September 2024	Retrieve data, extract insights, answer questions, and break down information silos by combining agents from different domains (e.g., supply chain, procurement, and finance)
Salesforce	Agentforce	September 2024	The catering, medical, retail, and financial industries can find customer orders on their own and modify them, replacing human jobs
Adobe	Agent Orchestrator	March, 2025	Allows companies to orchestrate AI agents to engage directly with customers and to support daily work across Adobe applications and third-party software.
ServiceNow	ServiceNow Agentic AI	September 2024	Integrate Agentic AI into the ServiceNow platform and unleash 24/7 productivity at scale across multiple use cases, including IT, customer service, procurement, HR, software development, and more.
Workday	Workday Illuminate	September 2024	4 AI agents : Recruiter, Expenses, Succession, and Workday Optimize Agents, simplify the daily work of human resources and finance departments
OpenAI	Code name Operator	January 2025	A computer can be used to take action on someone’s behalf, such as writing code or booking a trip
OpenAI	Canvas	December 2024	ChatGPT built-in Canvas can automatically trigger writing and programming, which is equivalent to an AI assistant for creators and programmers, allowing users to interactively collaborate with ChatGPT to complete a project.
Anthropic	Claude	October 2024	Using computers like humans, completing complex tasks in hundreds of steps, AI agents can be asked to book flights, fill out forms, conduct online research and submit expense reports in the future

Table 1: AI Agents launched by major companies

AI leaders’ progroess and opinion

Nvidia

Agent-based AI is regarded by Nvidia CEO Jensen Huang as the next big trend. Major technology giants are committed to building “AI Agents”. The large language model based on generative AI is the next main battlefield.

AI agents refer to AI that has the ability to make decisions and perform tasks autonomously in a specific environment under limited human supervision. After humans issue complex instructions (prompts), the AI agent understands human language, grasps changes in the environment, makes appropriate decisions, and can learn independently and adapt to the environment in response to changes to improve the performance of the next task.

Unlike AI chatbots, which are designed to converse with humans, AI agents receive tasks from the developer and complete them independently, possibly without interacting with other machines or humans at all.

On November 25, 2024, Jensen Huang, founder and CEO of Nvidia, attended the 2024 Hong Kong University of Science and Technology Honorary Doctorate Awarding Ceremony. Jensen Huang was awarded an honorary doctorate in engineering and delivered a speech, talking about the changes and prospects of artificial intelligence, and mentioned the two major trends of robots and AI Agents.

Regarding AI Agents, Jensen Huang, CEO of Nvidia, said: “Businesses and companies around the world are racing to adopt artificial intelligence to accelerate innovation and improve productivity. Soon, AI Agents will be integrated into every team of the company, including marketing, sales , supply chain, chip design, software development and other departments.

As shown in Table 1, as early as October 2023, Nvidia launched Eureka AI Agent, which used GPT-4 to generate a reward function and taught the robot to complete more than thirty complex tasks. For example, turning a pen quickly, opening drawers and cabinets, throwing and catching a ball. Especially for the pen-turning skill, it is very difficult to rely on humans to animate frame by frame. It surpasses human experts in more than 80% of tasks, increasing the average performance of robots to more than 50%.

Microsoft

In addition to February 2023, the company continues to launch a series of Copilot family products and integrates AI agent functions into Microsoft’s existing major products. It supports writing emails and reports, whiteboard brainstorming, OneNote planning, generating Excel formulas and charts, budget or accounting drafts, visual presentations and other important functions.

Microsoft also showed off an AI agent with generative AI reasoning capabilities in November 2024. The function of creating automated AI agents through Copilot Studio has been publicly previewed in November, using Copilot as an AI assistant to allow users to interact with AI agents.

Alphabet

Google, owned by Alphabet, officially launched an AI agent on December 11, 2024 that can be used to control the Chrome web browser, move the cursor on the screen, click buttons and fill out forms, allowing it to use and browse websites like humans. It can complete tasks such as research and shopping, and help people automate daily routine tasks completed through the web. Assist people with daily tasks such as shopping and booking flights.

Google launched the Gemini model on December 6, 2023, and expected to surpass its competitor OpenAI and its main product GPT-4 at the time. One year later, on December 11, 2023, Google announced that it was moving towards “Gemini 2.0”, moving towards an AI agent that can independently complete complex tasks.

OpenAI

OpenAI will launch the world’s first inference AI model “o1” in September 2024. It focuses on inference and will “think before answering questions.” After the AI conducts in-depth network research, it independently generates answers, surpassing existing conversational AI. generation function.

Through pre-training on large-scale data, OpenAI has significantly enhanced o1’s reasoning capabilities and is more suitable for handling difficult scientific and mathematical problems that cannot be answered by current commercial AI models. It has become a “smarter AI assistant” in formulating business plans and other strategies. .

Scheduled to be launched in January 2025, the AI Agent, codenamed Operator, can use a computer to take actions on someone’s behalf, such as writing code or booking a trip.

Anthropic

OpenAI’s biggest rival, AI start-up Anthropic, launched a new function called “tool use” in June 2024, which can be connected to any external API, data and tools selected by the user, thereby giving full play to the power of AI assistants and agents. Role. For example, the tool can analyze data, create personalized product recommendations based on users’ purchase history, and provide quick responses to customer inquiries. Make it easy for anyone to create an email assistant, a shoe-buying bot, or other personalized solutions.

AI agents will also be launched in October 2024, using computers like humans to complete complex tasks in dozens or even hundreds of steps. In the future, consumers can ask AI agents to book flights, fill out forms, conduct online research and submit expense reports. .

Amazon

Amazon Bedrock automates complex tasks such as advertising campaigns and processing insurance claims through custom instructions, orchestration, and monitoring.

Amazon is also building a super powerful AI chatbot code-named Metis, which can answer questions related to text and images in an intelligent conversational manner, provide corresponding links to information sources, recommend follow-up queries, and even generate images, and use a tool called ” Retrieval-augmented generation (RAG) AI technology obtains information beyond training data and generates the latest information.

Amazon aims to build Metis into an “AI agent” that can automatically perform complex tasks for users through its ability to understand, learn and analyze large amounts of data, such as providing the latest stock prices, planning vacation itineraries, controlling smart home devices, etc., and It is rumored that Amazon is also integrating Metis and its digital voice assistant Alexa.

Tag: CHGG

AI Agent will be the next wave of software industry