On November 30, 2022, OpenAI released ChatGPT almost as an afterthought. It was a polished version of technology that had been sitting in research labs for a while, dressed up in a simple chat box anyone could type into. The company expected a modest research preview. What they got was the fastest adoption of any consumer product in history up to that point. By some estimates it reached 100 million users in two months. For comparison, it took the telephone about seventy-five years to reach that many people.

This is part two of the story.

Part one ran seventy years, from a room-sized computer in 1956 up to the week before all of this. Part two covers about three and a half years. That alone should tell you something about the pace we are now living in.

Why it landed when it did. What made ChatGPT different was not that it was smarter than what came before. It was that you could just talk to it. No code, no special training, no manual. You typed a question in plain English and it answered in plain English. Suddenly your aunt, your accountant, and your child's teacher were all using the same tool, often for completely different reasons. People wrote cover letters with it, asked it to explain their medical results, and had it plan birthday parties. The technology existed. The on-ramp did not. That is the whole reason this date matters more than the quiet breakthroughs that led up to it.

A product that grows that fast does not go unnoticed. In early 2023 the competition arrived in a hurry. Google, which had quietly invented a lot of the underlying technology, scrambled to release its own chatbot. A company called Anthropic, started by former OpenAI researchers, put out an assistant named Claude. Meta took a different path and released its models for free, letting anyone download and build on them. Then in March 2023, OpenAI released GPT-4, a noticeably sharper model that could handle harder reasoning, score well on professional exams, and even look at images. For about a year, each new release felt like a genuine leap. People reopened questions they thought were settled about what software can and cannot do.

Then came the images. While the chatbots grabbed headlines, a parallel revolution was happening with images. Tools with names like Midjourney, DALL-E, and Stable Diffusion learned to turn a sentence into a picture. You typed "a corgi astronaut painted like a Renaissance portrait" and a few seconds later, there it was. Some of these tools were free and open for anyone to run. The results went from obviously fake to unsettlingly good in a matter of months. This was the first time a lot of people felt the ground move under them, because making images used to require either talent or money, and now it required a sentence. It also kicked off hard questions that are still unresolved, about who owns art that a machine learned from millions of human artists without asking. Courts, companies, and artists still do not agree on what "fair" looks like here.

It did not take long for the excitement to collide with reality. Writers, artists, and newspapers started noticing their work had been used to train these systems without permission or payment, and they began filing lawsuits. People discovered that chatbots confidently make things up, a problem the industry politely calls "hallucination," and a few folks got burned trusting answers that were flat wrong. Schools panicked about cheating, then slowly realized banning the tools was pointless. Governments started paying attention too. The European Union passed a sweeping AI law, and leaders around the world began arguing about how to handle a technology moving faster than anyone could regulate it. The honeymoon was short; the hard questions showed up almost immediately, and most of them are still open.

By late 2024, the labs hit a new gear. The first wave of chatbots answered instantly, blurting out the first thing that seemed right, which is part of why they got facts wrong. The newer "reasoning" models do something closer to working through a problem step by step before answering. They take longer, and they cost more to run, but they are dramatically better at math, logic, and anything that needs careful thought. That extra "thinking time" is quietly reshaping what people trust these systems to handle. If the early models were a clever student rushing through a test, these were the same student finally being told to show their work. For the first time, the machines were getting better not just by being bigger, but by being given room to think.

The shift happening right now is the biggest one yet, and it is harder to see because it is less flashy. The early tools answered your questions. The newer ones, often called "agents," try to actually do things for you. Instead of telling you how to book a trip, an agent attempts to book it. Instead of explaining how to fix a bug in your code, it goes in and fixes the code, then checks whether the fix worked. Instead of reminding you to follow up with a client, an agent drafts the email, sends it, and logs the note in your CRM. This is the leap from a tool that talks to a tool that acts, and it is genuinely useful and nerve-racking at the same time. As someone who has broken production code late at night, I can tell you: a system that not only suggests a fix but runs the tests for you lands very differently than a chatbot that just explains the error message. A system that can take actions on your behalf can also take the wrong ones, which is exactly why this is the part of the story still being figured out in real time.

So where does it actually stand. As of mid-2026, AI is woven into ordinary life in a way that would have sounded like science fiction four years ago. It writes and edits, generates images and video, powers customer service, helps doctors and lawyers and programmers, and sits inside the phone in your pocket. Hundreds of millions of people use it every week. Whole companies have been built on top of it, and plenty of older ones are scrambling to retrofit it into what they already do. If you walk into a random office in Fairhope or New York right now, odds are decent somebody there has a browser tab open with an AI tool they use every day. And it is still expensive, still makes mistakes, still raises legal and ethical questions nobody has fully answered.

Here is where this story goes next. Nobody actually knows. The people building these systems disagree with each other, loudly, about how far this goes and how fast. Some think we are a few years from machines that match human intelligence across the board. Others think we are hitting walls that more data and bigger computers will not solve. Both groups are smart, well-informed, and looking at the same evidence. That sort of disagreement is not a sign that no one knows anything; it is what real frontier work looks like.

Seventy years of buildup led to November 2022, and in the three and a half years since, AI went from a curiosity to something that might reshape how work itself functions. That's a lot of change in a short window. The field spent most of its life being five years away from everything, and it isn't five years away anymore. It's here, it's in your pocket, and the parts that matter most haven't happened yet.

LA-AI: Foley

Know someone who’d enjoy this?

Don’t miss the next issue.