There is something unusual about the way artificial intelligence is developing.
Humanity has always built tools to make difficult things easier.
The steam engine helped us produce mechanical power on a scale that human muscles simply couldn't match. Electricity then transformed how machines, factories, cities, and communication systems worked.
But AI is doing something different.
For the first time, we are building tools whose main purpose is not to make our bodies stronger or our machines more powerful.
It is trying to make our cognitive work faster.
And the speed at which this is happening is difficult to ignore.
From muscles to machines to intelligence
The Industrial Revolution did not happen overnight.
The steam engine required factories, railways, mines, transportation networks, and enormous amounts of physical infrastructure.
Electricity brought another transformation, but it also required power plants, cables, electrical systems, appliances, and entire cities to be redesigned around it.
AI has a very different advantage.
Its infrastructure is still enormous — data centers, chips, electricity and networks are absolutely necessary — but once a capable model exists, improvements can be distributed through software.
You don't need to build a new railway to make a model better.
You don't need to replace every machine in a factory.
Sometimes, you update the model.
And suddenly, millions of people can have access to a new capability.
That difference may be one of the most important things about AI.
We are only a few years into this
The timeline is also surprisingly short.
GPT-3 was released in 2020, but the moment that made generative AI a mainstream product came with ChatGPT on November 30, 2022.
GPT-4 followed in March 2023, already showing strong performance across professional and academic tasks and accepting both text and images as input.
Then the development accelerated.
AI went from answering questions to writing code.
From writing code to using tools.
From using tools to navigating websites.
From navigating websites to interacting with software.
And from doing individual tasks to completing sequences of tasks.
GPT-5, for example, was already designed to work as a coding collaborator, fixing bugs, editing codebases and performing longer chains of tool calls.
GPT-5.4 then pushed computer use further, allowing models to interact with computers through visual interfaces and perform real tasks across websites and software.
Now we have GPT-6 Astra.
And the difference is not simply that it gives better answers.
It is increasingly able to do things.
The prompt is becoming less important
Early generative AI had a very particular relationship with the user.
You had to explain what you wanted.
Then you had to correct the model.
Then you had to explain what it misunderstood.
Then you corrected it again.
A good result often depended on knowing how to write a good prompt.
Astra points toward something different.
OpenAI says the model can handle multi-step workflows across code, browsers and professional software. It can also fill in routine gaps when an instruction is incomplete and ask questions when missing information could change the result.
That sounds like a small improvement.
It isn't.
It changes the relationship between the human and the machine.
The question is slowly moving from:
“How do I tell AI exactly what to do?”
to:
“What do I want AI to accomplish?”
That is a much bigger change.
From answering to operating
Consider something as simple as a spreadsheet.
A traditional chatbot might explain how to create a financial model.
A more advanced system can actually work with the spreadsheet.
Astra is designed to create and modify documents, presentations and spreadsheets while following existing templates and styles.
The same idea extends to software.
Instead of asking:
“How do I fix this bug?”
you can increasingly ask an AI system to find the bug, modify the code, test the result and report back.
And then there are tasks that would have sounded ridiculous only a few years ago.
OpenAI demonstrates Astra working inside Blender, creating a 3D house and turning it into a walkable Unreal Engine scene.
That is not simply text generation.
The model is interacting with professional software to produce something that exists outside the conversation.
The chatbot is becoming an interface to other tools.
And then there is the problem of reality
AI-generated images and videos have created another interesting change.
There was a time when fake images were relatively easy to identify.
Then they became convincing.
Then video generation started producing scenes that could look surprisingly realistic.
OpenAI's Sora 2, for example, was designed specifically for more physically accurate and controllable video generation, with synchronized dialogue and sound.
This creates a strange situation.
The problem is no longer simply:
“Can AI create something that looks real?”
It increasingly can.
The problem becomes:
“How do I know whether something I am seeing is real?”
That is a very different challenge.
And it is another example of how quickly the technology is moving.
The most surprising example may be mathematics
Perhaps the clearest sign of where this could lead appeared only days ago.
On September 8, 2026, OpenAI announced that an internal AI system had produced a proposed solution to the Navier–Stokes existence and smoothness problem, one of the seven Millennium Prize Problems.
These equations describe the behavior of fluids, and the problem has remained unresolved for roughly 90 years.
OpenAI says its system produced a proof and a formalization in Lean. It also says the internal system was significantly more capable than GPT-6 Astra.
But there is an important detail.
We should not simply declare that AI has officially solved a Millennium Prize Problem.
The result still needs independent mathematical verification, and the announcement has already generated controversy around the use of previous research and questions of priority.
Still, even if the proof eventually requires corrections, the event itself is remarkable.
A machine is being used to attack a problem that humans have struggled with for decades.
And this raises a much bigger question.
What happens when intelligence becomes a tool?
The steam engine gave humans more physical power.
Electricity gave machines an enormous amount of flexibility and scale.
AI is beginning to give individuals access to something different: cognitive leverage.
A student can ask questions that would previously require searching through several books.
A programmer can generate and debug code.
A designer can create a 3D prototype.
A business can automate repetitive workflows.
A researcher can process enormous amounts of information.
A person with no experience in a professional tool can sometimes use natural language to operate it.
This doesn't mean expertise has become useless.
In fact, the opposite may be true.
As AI becomes more capable, knowing when its answer is wrong becomes more important.
Understanding the fundamentals becomes more important.
Being able to verify its work becomes more important.
The machine can increasingly do the execution.
Someone still needs to decide whether the execution makes sense.
Maybe this is why AI feels different
The steam engine changed what humans could physically do.
Electricity changed what machines could do.
AI is starting to change what individuals can think through and accomplish with the help of a machine.
And there is another difference.
The steam engine did not improve itself every few months.
Electric motors did not suddenly become better because someone uploaded a new version of their software.
AI systems can.
That creates a strange feedback loop:
better models → more useful tools → more users → more applications → more investment → better models.
The amount of time between one capability and the next keeps getting shorter.
ChatGPT became publicly available in late 2022.
It feels like a completely different technology today.
And GPT-6 Astra may already be another step toward a world where AI isn't simply answering us.
It is working alongside us.
We may not be watching another Industrial Revolution
Maybe calling AI the “next Industrial Revolution” isn't quite right.
It could be something different.
The Industrial Revolution multiplied our physical power.
The digital revolution multiplied our ability to store and transmit information.
AI may multiply our ability to turn information into action.
And that might be why its development feels so unusually fast.
The most important question may not be whether AI will replace programmers, designers, researchers, students, or other professionals.
It may be something much harder to answer:
What happens when the amount of work one person can accomplish is no longer limited by how quickly that person can personally perform every step?
We are still very early in finding out.
And if GPT-6 Astra is any indication, the next few years may be less about AI becoming a better chatbot...
and much more about AI becoming a general-purpose collaborator capable of doing the work itself.
That is probably the most important shift to watch.
Not because machines are suddenly becoming human.
But because, for the first time, humanity has built a tool that can increasingly participate in the process of producing knowledge, software, designs, research and decisions.
And unlike the steam engine or the electric motor, this tool is improving while we are still learning how to use it.
That may be what makes this revolution different.
Top comments (0)