This is StepbackThe weekly newsletter breaks down one basic story from the world of technology. More information about all things, AI, Follow Hayden field. Stepback appears in the inboxes of our subscribers at 8 am ET. Go to Stepback Here.
It all started with Jarvis so that Jarvis from Marvel’s films.
Well, maybe not start With the AI Iron Man assistant, but the fictitious system definitely helped the concept of AI agent. Whenever I interviewed people from the AI industry about Agentic AI, they often point to Jarvis as an example of an ideal AI tool in many ways – those that you know what you need before you ask, can analyze and find insights in immense seques of data and can offer strategic advice or a point of launch in some aspects of your business. People sometimes disagree about the exact definition of AI agent, but at its foundation it is a step beyond chatbot, because it is a system that can perform multi -stage tasks on your behalf without constant necessity of communication with you. He basically creates his own list of subtleties that he must complete to reach the preferred end goal. This fantasy is closer to being reality in many ways, but when it comes to the actual usefulness for a everyday user, there are many things that do not work – and maybe it will never work.
The term “agent AI” has been around for a long time, but yes especially He started popularity in the technology industry in 2023. It was a year concept AI agents; This term was on everyone’s lips when people tried to throw this idea and how to realize, but you have not seen many successful operate. The next year, 2024, was the year application – People really placed the code in the field and see what they can do. (At that time the answer was … not much. And filled with several error messages.)
I can indicate the noise around AI agents who become common to one specific announcement: in February 2024 Klarana, FinTech, company, he said that after a month his assistant AI (driven by Tech) successfully performed the work of 700 full -time customer service agents and the automated two -thirds of the company’s customer service. For months, these statistics appeared in almost every AI industry conversation, which I conducted.
Hype has never died, and in the following months, each general director of great technology seemed to harp at this time at each call for earnings. Amazon, Meta, Google, Microsoft and many other companies began to talk about their involvement in building useful and successful AI agents – and tried to put money where their lips are to happen.
The vision consisted in the fact that one day AI agent could do everything from travel booking to generate visualizations of business presentations. The perfect tool can even, say, find a good time and place to spend time with a group of your friends who work with all your calendars, food preferences and dietary restrictions – and then reserve a dinner reservation and create a calendar event for everyone.
Let’s talk about “AI coding” now: wearing Agentic Ai Industry. If you asked someone about real, successful, dissatisfied cases of operate for AI agents now And not conceptually in a not very distant future, they point to coding of artificial intelligence-it was almost the only specific thing that they could indicate. Many engineers operate AI agents for coding and they are considered objectively quite good. Good enough in Microsoft and Google, up to 30 percent The code is now written by AI agents. And in the case of startups such as Opeli and Anthropic, which smoke at high rates, one of their largest revenue generators are AI coding tools for corporate clients.
Therefore, until recently, AI coding was the main case of the operate of AI agents, but of course it is not subject to everyday consumer. Remember that the vision has always been an AI agent for “everyman”. And we are not there yet – but in 2025 we approached closer than ever before.
In October last year, Anthropic began the situation by introducing a “computer”, a tool that allowed Claude to operate a computer such as human power – browsing, searching, gaining access to various platforms and performing intricate tasks on behalf of the user. The general consensus was that the tool was a step forward for technology, but the reviews said that in practice it left much to be desired. Quick forward until January 2025, and Opennai was issued by the operator, his version of the same thing, and issued it as a tool for filling out forms, ordering foodstuffs, travel booking and creating memes. Once again, many users agreed that the tool was Buggy, leisurely and not always capable. But it was a significant step again. In the following month, Opeli issued Deep Research, an Agentic AI tool that could develop long research reports on any subject for the user, and which also reversed. Some people say that research reports were more impressive than content, but others were impressed. And then in July Opeli combined deep research and operator in one AI agent product: Agent Chatgpt. Was it better than most AI agency tools that appeared earlier? Absolutely. Was it still arduous to do work in practice? Absolutely.
So there is a long way to achieve this vision of the perfect AI agent, but at the same time we are technically closer than ever before. That is why technology companies put more and more money into agency artificial intelligence, by investing in additional calculations, research and development or talent. Google has recently hired a CEO, co -founder of Windsurf and some members of the research and development team, especially to lend a hand Google in exceeding AI agents. Companies such as anthropic and openai race in the ladder, rungs to introduce incremental features to put these agents in the hands of consumers. (For example, the anthropic has just announced the Chrome extension for Claude, which allows him to work in your browser.)
In fact, what will happen next, we see that coding of artificial intelligence is still improving (and unfortunately potentially replace many work Basic software engineers). We will also see that the products of agents suffered by consumers are improving, probably slowly, but for sure. And we will see how agents were increasingly applied to enterprise and governmental applications, especially since anthropic, Opeli and XAI debut AI platforms specific to the government in recent months.
In general, expect more false starts, starts and stopping as well as mergers and acquisitions as the AI agent’s competition increases (and the bubble noise will still be a balloon). One question that we will all have to ask during the months: what do we really want the conceptual “AI agent” so that it can do for us? Do we want them to replace only logistics, as well as more personal, human aspects of life (i.e. helping writing wedding toasts or notes for the supply of flowers)? And how well they lend a hand in logistics vs. Personal things? (Answer the latter: not very good at the moment.)
- In addition to the AI astronomical environmental cost – especially in the case of immense models that supply the efforts of the AI agent – there is an elephant in the room. And this is the idea that “wiser artificial intelligence that can do everything for you” is not always good, especially when people want to operate it to do … bad things. Things such as creating chemical, biological, radiological and nuclear weapons (CBRN). The best AI companies say that they are more and more worried about the risk of this. (Of course they don’t worry enough to stop building.)
- Let’s talk about the regulation of all this. Many people are afraid of the consequences of artificial intelligence, but many are not fully aware of the potential dangers that Uber-Helpphul, aiming AI agents create in the hands of evil actors, both in the United States and abroad (think: “habing vibe”, romance fraud and others). AI companies claim that they are ahead of the risk associated with the voluntary security they have implemented. But many others say that this may be an external intestinal checking.
0 Comments