Why AI Will Transform the World by 2030

In June 2025, trucks carrying what looked like wooden cabins drove deep into Russia. Hidden inside were Ukrainian drones. Then the roofs opened. 117 drones flew out and hit nuclear capable bombers parked as far away as Siberia. Drones are transforming modern warfare and their impact is just starting. In Operation Spiderweb, every drone still had a human pilot. Flying it remotely using Russia's own mobile network. The Ukrainian outlets did report that an AI trained on photos of old Soviet bombers from a Ukrainian museum helped the drones find the weak spots in the grounded bombers. So what happens when the drone doesn't need a pilot at all? Autonomous robot armies could be coming and it's likely that its first soldiers will fly. But for many of us, 2026 still feels pretty normal. Same office, same commute, same job. So how does the world go from our current everyday to a 2030 with robot armies? Almost four years ago, ChatGPT set off the current AI revolution. And we're about four years from 2030. So we can take today as the midpoint and explore the rest of the 2020s AI exponential. Now in the next four years, AI is increasingly going to change how we use our computers, do science, and it's going to enter the labor markets. Now how much that happens is based on the slope of the exponential. And the biggest question is going to be who ends up holding the most power. It doesn't start on the battlefield, it starts on the machine you're watching this on. Almost every office job is on a computer, and AIs are getting really good at using them. There's a test for this called OS World. Ordinary jobs on a real computer. Fix a spreadsheet, edit a slide deck, get something out of an inbox. In 2024, the best AIs finished about 12% of them. People finished about 72%. Today, the top models are above 85%. So the researchers built a harder version with jobs that take a person over an hour and a half and hundreds of steps to complete. The best AIs already score 78% on that one. The thing doing it is called an agent. AI agents aren't a new kind of robot, and I guess most of you watching already use them. Claude CoWork, Codex, OpenClaw, And the new ones from just last month, Meta's Muse and OpenAI's Dots. It's the same chatbot you already talked to put in a loop and give in tools. Engineers call it a react loop, short for reason and act. It looks at what's happening, often a screenshot, picks one action, takes the action, then looks at what happens and repeats. Click here, type this, run that, hundreds of times for every task. Now, one of the trends that really matters is how long of a job can an AI agent finish on its own? A research group called Meter tracks this. In 2023, AI could finish a coding task that takes a person about four minutes. By spring of 2026, it could finish a task that takes a person two full workdays. But there's a catch. On tasks that long, it only succeeds about half the time. However, the length of the task has been doubling about every four months. So even if it slows to doubling once a year, by 2030, AI will be doing tasks that could take a person six whole weeks. So if AI is this good, why does 2026 still feel kind of normal? Most offices haven't changed much yet. Last winter, a team of economists out of Stanford surveyed about 6,000 executives in the US, UK, Germany, and Australia. Most of their companies already use AI, but 9 in 10 said it hadn't changed their productivity or headcount. The ability is here, but the rollout is really just starting to pick up speed. So the coming years will be the really dramatic changes. So what changes first? Probably anything in a website or a portal. Paperwork, DMV forms, insurance claims, expense reports, and you'll be clicking through apps a lot less. Simply tell the AI what you want and the agent will do it. Using a computer will increasingly be less clicking and typing and a lot more talking. Programmers are already living this. I write almost no code by hand anymore. However, I am constantly making decisions and talking to the AI about architecture and goals. Many of our jobs could turn into being managers of AI agents. Here's a test you can run on your own job to see how AI agents will impact it. Is it on a computer? Can you describe what complete looks like? Can someone check if it's If you answered yes to all three, there's a good chance an agent can can probably do a lot of it before 2030. And of course, most jobs are a bundle of computer use and other tasks. A nurse's shift has charting in it, an electrician sends invoices, the computer part will likely be automated first. Now the danger to society isn't only layoffs, it's also that new jobs may never open. Stanford found that in the most AI-exposed jobs, 22 to 25-year-olds have fallen about 19% behind their peers in the least exposed jobs. The second danger is trust. In 2024, an employee at an engineering firm joined a video call with what looked like the company's CFO and several colleagues. Every one of them was actually a deep fake. The employee sent about $25 million. When anything on a screen can be faked, A lot of our current trust heuristics fall apart. Now picture scammers with thousands of agents targeting the most vulnerable people. Here's an interesting part. For a while, the least AI impacted jobs might be the ones done with our hands in the physical world. The agent that does your spreadsheet is much closer to being fully autonomous than a humanoid robot that does plumbing in your house. But that may not be true for as long as we assume. In September, OpenAI said 10,000 AI agents produced a proof for the Navier-Stokes millennium prize problem. And last week, OpenAI released 722 math papers. In many ways, math is the one place where AI is moving the fastest, and there's a simple reason why. AI improves the fastest whenever there's a cheap way to check if it was right, and math often has one. Write a proof in a program like Lean, and running it says yes or no. In contrast, most of biology doesn't have an easy way to check correctness. To check a drug, you often have to wait years to see who gets better. That's why AI is already doing new math, but we haven't yet cured cancer. Where there's a checker, the machine can guess, check, guess again, millions of times. A person won't try a million ideas that look dumb. We have pride in careers, whereas a machine has neither. So some of its most interesting finds may come from paths nobody bothered to try. People have started calling it alien math. Mathematicians say some of those 722 papers are so strange, and perhaps badly written, that even specialists can barely follow them. That's the weird future here, a proof can be right, and nobody can tell you why. Now let's put that AI loop in a wet lab, the kind with test tubes and chemicals. The AI picks the experiment, a robot arm runs it, and the result goes straight back to the AI, running around the clock. They're called self-driving labs, and they already exist. In medicine, AI's first big win is reading scans. In a Swedish trial of about 100,000 women, an AI that read breast scans helped doctors catch 29% more cancers, with no rise in false positives. So where are the cures? A new drug takes about 10 years of trials, and no drug designed mainly by AI has been approved yet. But think about why medicine works the way it does. A new drug can cost over a billion dollars to bring to market, so it only pays if a lot of people take it, or if a few people pay a fortune. Some rare disease treatments cost over 2 million dollars a patient. Either way, the drug is designed once, and everyone gets the same thing. Now let's say designing the drug gets cheaper. Then you can design drugs for individuals, targeted specifically to their unique illnesses. And that's already starting. In August 2026, Merck and Moderna said their skin cancer vaccine reached its main goal in a final stage trial. Every patient gets a different vaccine, designed by software from their own tumor. So the trial wasn't testing one specific drug, it was testing a recipe. And the untried paths show up here too. There's a disease called pulmonary fibrosis where your lungs slowly fill with scar tissue until you can't breathe. One company says its AI went looking for the protein that drives that scarring and found one that nobody had linked to the disease. Then it designed a pill to block that protein. Block the protein and the scarring slows down. That pill is now in its final stage trial. I don't think AI cures all cancers, but I think medicine could become far more targeted and certain diseases could be handled far better than today. Now the dark half. The same AI that helps design a drug could help someone design a disease. The big AI companies block this hard, sometimes so hard it gets in the way of researchers doing legitimate work. But those guardrails only exist on the big closed models. Open-weight models can be downloaded and run without safety filters. So the real risk isn't ChatGPT or Claude, it's a skilled person or group with open models and perhaps even an automated wet lab to run their experiments. Math was first because it all happens inside of a computer, but every day our labs get better at letting AI experiment in the real world. Physical labor might feel secure, AI can write an essay, but it can't fold a towel or fix fix your plumbing, but that's ultimately also going to change. In over 10 US cities, self-driving cars can already pick you up and drop you off somewhere. Waymo does about half a million rides a week, 10 times more than in spring 2024. Trucks are probably next. A human trucker has to stop and rest. A robot truck doesn't. In America, there's only about a few dozen working so far, but paying the driver is about 44% of what it costs to run a truck. However, there are many intricacies to automated freight. Especially in the last mile of the delivery. Now perhaps the most transformative technology we could ever build. General humanoid robots. There are already over 4 million robots in factories. But each one is a specialist. It does one job in one spot with the same motion over and over. Put it somewhere new and it's lost. The generation of robot that will truly change the world will be the generalists. They'll likely run the same kind of AI as ChatGPT. But instead of the next word, they'll predict their next move. And since they learn from basically all of humans' knowledge, they already know what most objects and things are, and are able to understand our real, physical world. And the bodies are getting better too. In Beijing, in April 2025, humanoids ran a half marathon. Most didn't finish. The winner was remote-controlled and took 2 hours and 40 minutes. One year later, the winner ran it on its own in about 50 minutes. Faster than any human ever has. And this isn't just demos. A company called Figure AI says its humanoids spent 11 months in a BMW assembly line in South Carolina, working up to 10 hour shifts and loading parts for more than 30,000 cars. The parts on the robots that broke the most were the forearms, wrists, and hands are still one of the hardest parts. A robot is a brain and a body. Teach one robot's brain something and you can copy it into a million others. But every body part has to be built in a factory. And by one count, more than 97% of the humanoids shipped in the first half of 2026 came from Chinese companies. Robots will join our world in the order that they are viable. The sky will have more and more drones, and our roads more self-driving cars than warehouses and factory lines. And finally, near the end, we will get our own generalized personal robots that can do all sorts of tasks for us, like plumbing. That is, if we can afford them. So by 2030, my guess is that robotaxis are normal in most cities, and humanoid robots ship in the hundreds of thousands a year, yet still mostly in warehouses and factories. Most physical jobs likely still have people doing them. But for the 4 million Americans that drive for a living, trucks, deliveries, taxis, they may find themselves in extreme competition with robots. Now who's buying all these? A business buys a robot only if it's cheaper than a worker. A military buys a drone if it's cheaper than a missile or than losing a soldier, and that bar has already been cleared. In Ukraine, a drone that costs a few hundred dollars can take out a tank worth millions. By 2024, a NATO official said drones were behind about two-thirds of the Russian tanks destroyed in Ukraine. So military force gets automated before work does. That's why the first robot armies won't wait for generalized humanoids. They're already flying today. The only difference is that for now, people are still at the controls. So let's put it all together and run the clock forward. This is my best guess, year by year. In 2027, agents become a much more common way to use a computer. Instead of clicking through apps, you just tell the AI what you want, and it goes out and does it for you. In 2028, AI starts taking on tasks that would normally take a person a week or more. Automated wet labs run experiments around the clock. And every major military is rebuilding itself around drones. Millions of them, every size, in the air, on the ground, and at sea. In 2029, robot trucks become common on the highways, with humans perhaps still handling the last mile. Humanoids move out of pilot programs and onto factory floors. Lots of medical drug research shifts to AI, and many of the drugs entering trials have AI somewhere in their design, and the first wave of them reaches final stage trials. And by 2030, AI can take on tasks that would normally take a person multiple months. The first AI-designed drugs start to reach patients, and the robot armies are now here. Millions of cheap drones, robots on wheels and tracks, and even some humanoids. Perhaps most importantly, there are fewer and fewer people at the controls. For all of history, projecting power has taken people. To run a country, fight a war, or enforce a law, you needed soldiers, workers, and officials willing to go along with it. That need kept power in check, and machines could change that. The fewer people it takes, the fewer people have to agree. So who holds power? Right now, a handful of companies and a couple of governments. America has the chips and China has the bodies and most of the parts. I'm genuinely excited about a lot of this. Paperwork disappearing, cancers caught earlier, and perhaps cured medicine that's personalized. Many of these machines are truly going to work. However, the bigger question is who they work for and what are their intentions. So what do you think? Is humanity heading in a good direction in the next four years? And if you got this far, drop a like and subscribe to the channel. More videos to come.