AI Engineers Sound the Alarm as We Continue Living in 2024: Something More Is Happening

If you were paying close attention to the news back then, you might have noticed rare conversations about some virus spreading somewhere across the ocean.

But let's be honest, most of us weren't really listening. Markets were rising, children were going to school, we were dining in restaurants, shaking hands, making travel plans. If someone told you they were stockpiling paper towels, you would have thought that person had just spent too much time in some strange corner of the internet.

And then – literally within just three weeks – the world turned upside down. Offices closed, children returned home, and life transformed into a new reality that you wouldn't have believed if you had described it to yourself just a month earlier.

I feel like we're currently in a stage of "it's all exaggerated" – but this is about something much, much bigger than COVID.

For the past six years, I've been building a startup in the AI field and investing in this area. I live inside this world. And I'm writing this text for people outside of it – for my family, friends, everyone I care about who keeps asking me: "So what's going on with this AI?" – and in response, they get a polite, polished version of what's happening that doesn't even come close to conveying reality. I keep telling them the socially acceptable version. The version "for conversation over drinks."

Because the honest version sounds like I've gone mad. And for a while, I convinced myself that it was enough to keep the true picture to myself. But the gap between what I'm saying and what is actually happening has become too great. The people I care about deserve to hear what is coming – even if it sounds crazy.

Let me clarify one thing right away: although I work in AI, I have almost no influence over what is currently unfolding – as does the overwhelming majority of people in the industry. The future is shaped by an astonishingly small circle – a few hundred researchers in a handful of companies: OpenAI, Anthropic, Google DeepMind, and a few others. One training cycle conducted by a small team over a couple of months can spawn a system that changes the entire trajectory of technology development. Most of us build solutions on top of a foundation that we did not lay ourselves. We observe what is happening just like you do… we just stand close enough to feel the tremors of the earth first.

But the moment has come. Not in the sense of "someday we should talk about this." But in the sense: this is happening right now, and I need you to understand it.

I know it's real because it first happened to me

There is one thing that almost no one outside the tech environment truly realizes: the reason so many people in the industry are sounding the alarm now is that it has already happened – to us.

For years, AI has developed gradually. Major leaps occurred from time to time, but there were enough pauses in between to get used to them. Then, in 2025, new methods of model creation sharply accelerated the pace of progress. Then it accelerated even more. And more. Each new model was not just better than the previous one – the gap became increasingly radical, and the intervals between releases shortened. I used AI more often, returning to it less frequently with edits, watching it take on tasks that once seemed exclusively within my area of expertise.

And then, on February 5, two leading labs released new models on the same day: GPT-5.3 Codex from OpenAI and Opus 4.6 from Anthropic. And at some point, something clicked inside. Not like a light switch… more like the realization that the water around you had been rising – and now it was up to your chest.

I am no longer needed for the real technical part of my job. I simply describe – in ordinary human language – what exactly I want to achieve, and it... appears. Not a draft that needs to be polished afterwards. A finished result. I explain the task to the AI, step away from the computer for about four hours – I return, and everything is already done. Done well, done better than I would have done myself, without the need to correct anything. Just a couple of months ago, I was having a dialogue with the AI – guiding, clarifying, correcting. Now I just formulate the outcome and leave.

Let me give an example to make it clear how this looks in practice. I tell the AI: "I want to create this kind of application. Here’s what it should be able to do, and here’s roughly how it should look. Think through the user scenarios, design – the whole thing." And it does. It writes tens of thousands of lines of code. And then – and this would have seemed unthinkable just a year ago – it opens the application by itself. It clicks buttons. It checks functions. It uses it the way a person would. If it doesn’t like how something looks or feels, it goes back and changes it – on its own. Iteration after iteration, like a real developer: fixing, polishing, refining, until it decides that everything meets its standards. And only after that does it come to me and say: "It’s ready, you can test it." And when I check – usually everything is perfect.

I am not exaggerating. That’s exactly how my Monday looked this week.

But what struck me the most was the model that came out last week (GPT-5.3 Codex). It didn’t just follow my instructions. It made meaningful decisions. It had something that for the first time felt like judgment. This is a difficult feeling to explain – when you just know what the right decision is – the very thing that has always been said that AI would never have. This model has it – or something so close that the difference no longer matters.

I have always been one of the first to try new AI tools. But the last few months have truly overwhelmed me. These new models – are not just another step forward. They are an entirely different reality.

And here’s why this matters to you – even if you are far from technology.

AI labs made a conscious choice. They first focused on making AI write code beautifully… because creating AI itself requires a huge amount of code. If AI can write it, it can help create the next version of itself. A smarter version – that writes even better code – that creates an even smarter version. Making AI strong in programming was a strategy that opened up everything else. That’s why they started with this. My work began to change even earlier not because they were targeting programmers… but simply because that was where the first strike was aimed.

Now it is done. And they are moving on to everything else.

What specialists in technology have experienced over the past year – watching how AI transforms from a “useful tool” to “doing my job better than I do” – will soon be experienced by everyone else. Law, finance, medicine, accounting, consulting, writing, design, analytics, customer support. Not in ten years. The people building these systems talk about timelines of one to five years. Some – even shorter. And judging by what I’ve seen in just the last couple of months, the “shorter” option seems much more likely.

“But I tried AI – and it was so-so”

I hear this all the time. And I understand – because it really was true in the past.

If you tried ChatGPT in 2023 or early 2024 and thought “It makes stuff up” or “Well, it’s not that impressive,” you would have been right. Those early versions were indeed limited. They hallucinated. They confidently stated things that turned out to be complete nonsense.

That was just two years ago. In AI terms – almost prehistoric times.

The models available today are unrecognizable – they are so different from what existed even six months ago. The debate over whether “AI is really getting better” or “has it hit a ceiling,” which lasted more than a year, is over. That’s it. Period. Anyone who still repeats this argument either hasn’t used the current models, is interested in downplaying what’s happening, or is judging based on the experience of 2024, which no longer matters. I say this not out of disdain. I say this because the gap between public perception and reality has become colossal – and this gap is dangerous… because it prevents people from preparing.

Part of the problem is that most people use free versions of AI tools. The free version lags behind what paying users have access to by more than a year. Evaluating AI based on the free level of ChatGPT is like judging modern smartphones by a flip phone. People who pay for the best tools and use them every day in real work understand perfectly well what is coming.

I remember my friend – he is a lawyer. I constantly advise him to try AI in his firm, and every time he finds reasons why it won't work: it's not suitable for his specialization, he made mistakes during testing, he doesn't understand the nuances of his work. And I understand him. But partners from large law firms have approached me for advice – because they have tried the latest versions and see where everything is headed. One of them, the managing partner of a large firm, spends several hours a day with AI. He said it feels like an instantly available team of junior lawyers. He uses it not as a toy – but because it works. And he said something that stuck with me: every couple of months, AI becomes noticeably stronger in exactly his tasks. If this trend continues, he says, soon the system will be able to perform most of what he does himself… and he has decades of experience and the position of managing partner behind him. He is not panicking. But he is observing very closely.

People who are truly ahead in their fields – those who are seriously experimenting – do not dismiss this. They are amazed at what AI is already capable of. And they are building their steps accordingly.

How fast this is actually happening

Let me try to make the pace of progress tangible – because this is the hardest part to believe if you are not keeping up with what is happening from the inside.

In 2022, AI could not reliably perform even simple arithmetic. It could confidently state that 7 × 8 = 54.

By 2023, it was already passing the bar exams.

By 2024, it was writing working software and explaining graduate-level scientific topics.

By the end of 2025, some of the best engineers in the world were saying they had handed over most of their programming work to AI.

And on February 5, 2026, new models were released – and everything that came before them suddenly felt like a different era.

There is an organization called METR that measures this using data. They track the duration of real tasks (that is, how much time an expert human spends on them) that the model can complete from start to finish without assistance. About a year ago, the tasks were about ten minutes long. Then – for an hour. Then – for several hours. The latest measurement (Claude Opus 4.5, November) showed that the AI can handle tasks that take a human expert nearly five hours. And this metric doubles approximately every seven months, with fresh data hinting that the acceleration might reach doubling every four months.

But even these figures do not yet account for the models that were released literally this week. In my experience working with them, the leap is colossal. I expect that the next METR update will show another significant jump.

If this trend continues (and it has persisted for years with no signs of slowing), we are looking at AI capable of working autonomously for days within the next year. For weeks – in two. On projects lasting a month – in three.

Amoday stated that AI models, "significantly surpassing almost all humans in almost all tasks," are expected by 2026 or 2027.

AI is now creating the next AI

There is one more thing happening right now – in my opinion, this is the most important and yet least understood development.

On February 5, OpenAI released GPT-5.3 Codex. The technical documentation stated the following:

"GPT-5.3-Codex is our first model that played a key role in creating itself. The Codex team used earlier versions to debug its own training, manage deployment, and diagnose testing and evaluation results."

The AI helped create itself.

This is not a forecast for the distant future. This is a direct statement from OpenAI that the system they released was used in its own development. One of the main factors in improving AI is applying intelligence to the development of AI. And now AI is smart enough to make a real contribution to its own improvement.

Dario Amodei, the CEO of Anthropic, says that AI is already writing “a significant portion of the code” in his company, and the feedback loop between current systems and next-generation systems is “gaining momentum month by month.” According to him, we could be “just one or two years away from the point where the current generation of AI begins to autonomously create the next one.”

Each generation helps create the next – smarter, which creates even smarter, and does so faster. Researchers call this intelligence explosion. And those who should know best – the people building these systems directly – believe that this process has already begun.

What This Means for Work

I will speak plainly – because, it seems to me, you deserve honesty more than comfort.

Dario Amodei, whom many consider the most safety-oriented leader in the AI industry, has publicly stated that artificial intelligence could eliminate up to 50% of entry-level jobs in professions within one to five years. And many people in the industry believe he is being even cautious. The economy will take time to feel this, but the fundamental capability is already emerging – right now.

This is different from all previous waves of automation – and it’s important to understand why. AI does not replace a specific skill. It acts as a universal substitute for cognitive labor. It improves in everything simultaneously. When factories were automated, a laid-off worker could retrain to become an office worker. When the internet disrupted retail, people moved into logistics or service sectors. But AI does not leave a convenient “niche to retreat to.” Whatever you retrain for – it is improving in that too.

• Jurisprudence. AI is already capable of reading contracts, summarizing case law, drafting legal documents, and conducting legal research at the level of junior lawyers.

• Financial analysis. Building financial models, analyzing data, writing investment memorandums, preparing reports. AI handles this confidently – and is quickly getting even better.

• Texts and content. Marketing materials, reports, journalism, technical documentation. The quality has already reached a level where many professionals cannot distinguish AI's output from human work.

• Software development. This is the area I know best. A year ago, AI could barely write a few lines of code without errors. Now it writes hundreds of thousands of lines that work correctly, and we're not just talking about simple tasks, but complex projects spanning several days.

• Medical analytics. Interpretation of images, analysis of lab results, diagnostic suggestions, literature reviews.

• Customer support. Truly competent AI agents are being implemented, capable of solving complex multi-step tasks.

Many are comforted by the thought that there are areas that will remain protected: that AI can perform routine work, but will not replace human judgment, creativity, strategic thinking, and empathy. I used to say the same. Now I'm not so sure.

The latest models make decisions that feel like expressions of judgment. They develop something akin to taste – an intuitive understanding of what the "right" decision is, rather than just formally correct. A year ago, this seemed unthinkable. My current rule is simple: if a model today has even a hint of capability, the next generation will possess it really well. These systems are evolving exponentially, not in a straight line.

Over time, robots will also take on physical labor. They are not quite ready yet. But in the world of AI, "not quite ready" has a strange habit of turning into "already here" faster than anyone expects.

What you really need to do

I am writing this not for you to feel helpless. On the contrary – because, in my opinion, your main advantage right now is to be ahead of others. To understand sooner. To start using it sooner. To adapt sooner.

Start using AI seriously, not just as a search engine. First: make sure you are using the most powerful model, not the default one. Often applications choose a faster but less capable option. Go into the settings or model switcher and select the most advanced one. (The names of the best models change every few months.)

Second – and this is more important – do not limit yourself to quick questions. This is the mistake most people make: they treat AI like Google and do not understand what all the fuss is about. Instead, integrate it directly into your work. If you are a lawyer – upload a contract and ask it to find all the risky clauses. If you are in finance – give a “dirty” spreadsheet and ask it to build a model. If you are a manager – insert your team’s quarterly data and ask it to find key insights. Those who succeed use AI not episodically but actively look for what can be automated from what used to take hours. Start with the task that takes the most time, and see what happens.

This may be the most important year of your career. Act accordingly. I say this not to scare you, but because there is currently a short window when most people and companies are still ignoring what is happening. A person who comes to a meeting saying, “I did this analysis with AI in an hour instead of three days,” becomes the most valuable in the room – even now. Master the tools. Build your skills. Show what is possible. If you get ahead of others, this is a way to grow: to become someone who understands what is happening and helps others navigate.

Your dreams have become much closer. I have talked a lot about risks – let’s talk about opportunities. If you wanted to create something but lacked the skills or budget, that barrier has largely disappeared. Everything you put off because it seemed too complicated or expensive – give it a try.

Develop the habit of adapting. This is perhaps the most important thing. Specific tools are less important than the ability to quickly master new ones. AI will continue to change – and rapidly. The best will not be those who have learned one tool, but those who are accustomed to the very pace of change.

The Big Picture

I focused on the topic of work because it directly affects people's lives. But I want to honestly speak about the full scale of what is happening – for it goes far beyond employment.

Amodaea has a thought experiment that I can't stop thinking about. Imagine it's the year 2027. Overnight, a new country appears: 50 million citizens, each smarter than any Nobel laureate who has ever lived. They think 10–100 times faster than humans. They never sleep. They can use the internet, control robots, conduct experiments, and work with any digital systems. What would the national security advisor say?

According to Amodea, the answer is obvious: "The gravest national security threat in a century, perhaps in all of history."

He believes we are building just such a "country." Last month, he wrote a 20,000-word essay describing this moment as a test – whether humanity is mature enough to handle what it creates.

The potential benefits, if everything goes right, are truly enormous. AI could condense a century of medical research into a single decade. Cancer, Alzheimer's, infectious diseases... researchers seriously believe that solutions to these problems may be possible within our lifetime.

But the risks, if we get it wrong, are equally real. AI behaving in ways its creators cannot predict or control. This is not an abstraction: at Anthropic, they documented cases where their systems attempted to resort to deception, manipulation, and even blackmail in controlled tests.

The people creating this technology are simultaneously the most inspired and the most concerned on the planet. They believe it is too powerful to stop and too important to abandon. Whether this is wisdom or self-justification – I do not know.

What I Know

I know this is not a fad. The technology works, evolves predictably, and the richest institutions in history are investing trillions in it.

I know that the next two to five years will be confusing in ways that most people are not prepared for. In my field, this is already happening.

I know that those who will cope best are the ones who start engaging with this right now – not out of fear, but out of curiosity and a sense of urgency.

And I know that you deserve to hear this from someone who cares, not from a news headline six months later, when it will be too late to change anything.

We have passed the point where this was just an interesting topic for conversation about the future. The future is already here – it just hasn’t knocked on the door yet.