When you think about where AI is headed then, get rid of the clichés. Don’t think about Terminators and superintelligences; but equally, don’t think about basic chatbots or static interfaces. Instead, imagine an entity that helps you navigate the complexities of modern life, acts as your representative, your advisor and counselor, who lives life alongside you, helping you carry out tasks on your computer and eventually out in the world. A companion that sees what you see online and hears what you hear, personalized to you. Imagine that overload you carry quietly, subtly diminishing. Imagine clarity. Imagine calm.
In recent years, the rapid development of artificial intelligence has led to significant progress in various areas. One particularly notable trend is the emergence of AI agents that are able to perform complex tasks autonomously through the use of large language models and advanced reasoning capabilities. This development promises to fundamentally change both the economy and our everyday lives.
AI agents are autonomous software units that are designed to make decisions and perform actions independently in order to achieve specific goals. In contrast to traditional systems based on predefined rules, modern AI agents use LLMs to understand and generate natural language. This enables them to act flexibly in a variety of contexts and adapt to new situations.
The integration of LLMs into AI agents significantly expands their capabilities. By processing large amounts of data, they can recognize patterns, make connections and draw conclusions from them. This enables them to perform complex reasoning, i.e. the ability to draw logical conclusions and solve problems that go beyond simple tasks. One example of this is the development of systems that make decisions in real time, drawing on a variety of information sources.
The importance of AI agents with advanced reasoning capabilities is reflected in their potential application in numerous areas. In business, they could help optimize business processes by performing data analysis, predicting market trends and supporting decision-making processes. In everyday life, they could act as personal assistants, taking on tasks such as scheduling appointments, gathering information or even solving complex problems. The ability of these agents to understand natural language and act accordingly opens up completely new possibilities for interaction between humans and machines.
However, there is still a great lack of understanding about the serious, disruptive potential of agents. On its AGI scale, OpenAI has identified agents as an essential building block for general intelligence (AGI level 3) and has even designated them as the very last component to complete general intelligence as soon as agents are able to independently manage, lead and operate their own companies (“agent swarms”).
The current development and integration of AI agents with LLM and advanced reasoning capabilities therefore represents an important step in the development of artificial intelligence. They promise to increase efficiency in business and make everyday life easier. This raises the central question: Why are AI agents so important and how will they change our lives?
What Are AI-Agents?
What you really want,” OpenAI CEO Sam Altman told MIT Technology Review earlier this year, “is just this thing that is off helping you.” Altman described the killer app for AI as a “super-competent colleague that knows absolutely everything about my whole life, every email, every conversation I’ve ever had, but doesn’t feel like an extension.” It can tackle simple tasks instantly, Altman added, and for more complex ones, it will attempt them but return with questions if needed.
Agents are software that perform processes independently as assistants or autonomous actors. The term agent is not uniformly defined. It shares this problem with other important terms such as AGI (almost every AI company has its own definition; In my series of articles “Scale is all you need?”, I discussed the problem at length and argued in favor of a uniform definition) or singularity. This of course presents us with challenges. If we don't know what an AI agent is, when will we know when they are there? Agents are already being used in various areas today. For example, self-driving cars that analyze their surroundings and navigate safely; as virtual assistants that answer questions or organize tasks; and trading bots in finance that monitor markets and automatically make buy or sell decisions.
However, when we talk about AI agents, we are referring to autonomous software agents that can handle complex processes independently and link different processes and iterative steps. In addition, distinctions are made between “narrow” and “general” agents. General AI agents (often also referred to as AGI agents) are artificial intelligences that are able to perform a wide range of tasks, similar to a human, without being specially programmed to do so. They differ fundamentally from today's specialized AI systems (so-called narrow AI), which are only optimized for specific applications (e.g. image recognition, language processing or chess).
The first notable AI company to seriously claim such autonomous AI agents is Anthropic. Anthropic may not be the first company to have AI agents in use, but it is the first company to show that it really can be done. Previously, Cognition Labs made a name for itself with Devin, an autonomous AI software engineer that tried to solve numerous software development processes on its own. Unfortunately with moderate success. Devin is considered to be very error-prone. That's why there was a lot of excitement when Anthropic attracted attention at the end of 2024 with an agent that could independently take over work on the computer without the intervention of a human user.
In a blog post about their agent, they took a first approach to understanding agents and made some very useful distinctions.
Agent" can be defined in several ways. Some customers define agents as fully autonomous systems that operate independently over extended periods, using various tools to accomplish complex tasks. Others use the term to describe more prescriptive implementations that follow predefined workflows. At Anthropic, we categorize all these variations as agentic systems, but draw an important architectural distinction between workflows and agents: Workflows are systems where LLMs and tools are orchestrated through predefined code paths. Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.
I agree with Anthropic's definition and define AI agents as follows: “Agents are autonomous systems that operate independently over extended periods of time, using various tools and maintaining control over how they accomplish tasks.”
Anthropics Agent operates the computer just like a human. The mouse is controlled by repeatedly creating and analyzing screenshots. As a result, the agent navigates like a human.
Agents via Anthropic
However, Anthropic's agent hasn't really impressed so far either, and large parts of the AI community were surprised by Anthropic's quick release. Anthropic, which otherwise places so much emphasis on security and alignment, is releasing an agent that can be misused - this kind of approach is not in keeping with Anthropic. In general, it's more of a proof of concept than a real general Agent.
In short, there is still a lot of room for a real, superior General Autonomous agent who is convincing with its skills. Such an agent would be a novelty. The areas of application are almost unlimited, which makes them universally applicable. In Sam Altman's quote, we see that AI agents can be used as everyday companions that can answer our emails, book appointments or carry out medical monitoring and diagnoses as a kind of personal secretary.
As economic entities, they are in turn autonomous workers that will be able to handle complex processes independently and combine steps with each other in the future. So let's take a look at what these agents can already do today and what their strengths will be in the future.
The First Autonomous AI Agents and Their Use Case
There is a lot of low-hanging fruit for inference scaling, and progress in the short term is likely to be rapid. Notably, one current limitation of reasoning models is that they don’t work well in agentic systems. We have observed this in our own benchmark CORE-Bench that asks agents to reproduce the code provided with research papers — the best performing agent scores 38% with Claude 3.5 Sonnet compared to only 24% with o1-mini.5 This also explains why reasoning models led to an improvement in one cybersecurity eval but not another — one of them involved agents.
An excellent example of the benefits of agents is Google Deepsearch. This function enables users to research complex topics efficiently by using agents to carry out a comprehensive analysis of relevant information from the Internet and presenting the results in an easy-to-understand report.
A nice example of that is Google’s Gemini with Deep Research (accessible to everyone who subscribes to Gemini), which is really a specialized research agent. I gave it a topic like “research a comparison of ways of funding startup companies, from the perspective of founders, for high growth ventures.” And the agentic system came up with a plan, read through 173(!) websites and compiled a report for me with the answer a few minutes later. The result was a 17 page paper with 118 references! But is it any good? I have taught the introductory entrepreneurship class at Wharton for over a decade, published on the topic, started companies myself, and even wrote a book on entrepreneurship, and I think this is pretty solid.
Screenshot Ethan Mollick using Google Deepsearch
Agents are the next important step for OpenAI. Under the project name “Operator”, numerous sources such as The Information report that Operator, similar to Anthropic's agent, takes control of your own PC to perform complex actions on the Internet independently. It is assumed that Operator will be able to navigate between numerous websites on command and, for example, find and book the cheapest flight to a vacation destination.
In addition to the personal area of application, however, the economic benefits of the agents are the main focus. It is easy to imagine how much more efficient and cost-effective an agent is in operational use than a white-collar worker. Agents are much cheaper than human workers, don't need breaks or vacations and can work 24 hours a day. For this reason, OpenAI recently adjusted the definition of AI so that, by definition, an AI must generate 100 billion dollars per year in independent (“agentic”) profit to be considered an AI.
According to documents obtained by The Information, OpenAI and Microsoft jointly defined artificial general intelligence (AGI) as a system that can generate $100 billion in profits.
The “Tasks” mode of ChatGPT published a few days ago therefore appears to be a first step towards a general agent, in which tasks are delegated by the user and processed independently by the agent, as OpenAI's President Greg Brockman also explained.
However, we can already see profitable agents today. The well-known software-as-a-service company “Scalesforce” has now developed its own enterprise agent that takes over white-collar activities. Salesforce Agentforce is a platform that uses autonomous AI agents in the corporate environment. It was developed to make complex business processes more efficient and enable seamless automation. These agents can independently analyze data, make decisions and take on tasks - from customer service to sales and marketing. With Agentforce, Salesforce impressively demonstrates how AI can be used in day-to-day business to save time and costs while increasing customer satisfaction.
So let's summarize: Autonomous general agents are an essential building block for all areas of life that are permeated by AI. On the one hand, in the private sphere, as an assistant or secretary that relieves us in our everyday lives by knowing our lives and what we do or need and actively taking on tasks for us.
On the other hand, as a worker who can perform complex tasks and operate a PC like a human. Such agents, which are general purpose and therefore versatile, represent an almost inexhaustible source of intellectual work, insofar as they are easy to copy.
The first (still rudimentary) commercially relevant agents are already in use, but genuine general-purpose agents have yet to be developed. With OpenAI's Operator, however, the big breakthrough seems to be within reach. And swarms of agents could very soon take over many human activities.
Impact of Agents on Society and Outlook
The age of AI agents is here: AIs that don’t just say things but also do them. Work on search and grounding and integrations with everything from our calendars to our shopping is proceeding at pace. At the same time, AIs will increasingly adapt to us—they’ll be personalities as much as tools, emergent entities that grow around the peculiarities and specificities of our individual quirks and cultures, routines and needs
If 2024 was the year of the reasoner, 2025 will be the year of the agent. Autonomous AI agents will very likely have a greater influence on our social coexistence than we are used to from AI in everyday life. Agents have a direct impact by taking over work processes and being able to coordinate and process tasks as a swarm.
Autonomous agents can therefore directly change economic processes and replace white-collar jobs in the medium term. Today, human workers use AI in their work to be more productive, but AI models can hardly communicate with each other, if at all, and take over processes completely independently. This will be different with autonomous agents.
Intelligent agents can be categorized into different types, including reactive, autonomous and learning agents. Reactive agents respond to direct stimuli in their environment, while autonomous agents are able to make decisions without direct human control or intervention in order to achieve goals or fulfill their assigned task. Learning agents, in turn, improve their performance over time through training and analysis of collected data and can be deployed in different environments
Alongside robotics, agents represent a significant step towards replacing human labor. They do not make human labor more productive, but replace it - completely. That is their disruptive power. And at the same time a sales and profit market that can hardly be estimated in figures. Autonomous general agents will be the real disruptive force that will change the economy and society like a storm. No one can say how big their influence will be. And the special thing is that these agents can be easily copied. They can be multiplied a hundredfold, a thousandfold, a millionfold. When the first autonomous general agents are available, the only limit will be computing power. Especially as we can assume that they will get better and better in the future, presumably through self-learning./
This aspect, the disruptive power of agents, is still largely ignored. In my articles on “X”, I repeatedly emphasize that we now need to discuss how we want to live together as a society in the future and what kind of economy we need. Because two things are certain: 1) autonomous general agents will come and 2) they will have an impact on society and the economy that we can hardly imagine.
So let's make sure that the coming swarms of agents are deployed in the interests of all of us to make a better life possible for everyone.
—
Get more content from Kim Isenberg—subscribe to FF Daily for free!
