We are transitioning to a transmedium culture. Just like how so much was lost when we left our oral roots, I do believe we are going to lose something important as we move away from being a written culture.

Oral cultures could remember long stories and pass them down nearly perfectly for generations. I believe that written cultures have a thoroughness to them that we may very well be at risk of losing.

You might be asking what I’m talking about, and why I think there’s a transition coming—or happening right now. The fact is that we can speak much faster than we can write, and we can read much faster than we can listen. The bulk of modern communication has become asynchronous, so people are going to opt for the medium that they are fastest in—and that is most convenient to them—for whichever action they are doing: input or output.

Increasingly, people are using voice transcription so that they can speak every thought they have while AIs write them down. I know that I find myself frequently giving prompts to AI through voice commands that are transcribed into text. But I do notice that when I write a prompt by hand, it is often much more thorough. I’m forced to think much more carefully about exactly what I’m saying and how I’m saying it.

While this same thoughtfulness can likely be cultivated in spoken transcriptions, there is a certain texture to writing something—the act of seeing it on your screen as you’re writing it, and reflecting on it slowly, as you read your words over and over again with each additional word you put onto the screen.

I think it is effectively inevitable that we will lose some thoroughness and some carefulness, and that is just a trade-off that we are likely going to have to accept.

I do think it’s important to get used to this now, because it is preparation for the next transition: the transition to a telepathic society. Every big tech company now is focusing on brain-computer interfaces. Telepathy is going to be the highest-bandwidth way for people to communicate with AIs and each other.

It’s going to be very primitive at first. It’s probably just going to take your direct thoughts and convert them into text that is broadcast to your recipients—and then beamed into their heads as a voice reading those texts aloud.

But many people think in different ways about different things. Some people think in shapes, some people think in images, some people think in sound and music. And I believe that we will be able to directly communicate those different modalities to each other in real time. Once we go there, there really is no going back.

Even what we are doing now—voicing our direct thoughts, just as I am voicing my direct thoughts into a microphone that is transcribing this very article right now—will be looked back upon as relatively quaint.

I know there are many of you who fear this transition to telepathy. You are worried about the privacy of your thoughts. You are also worried about the impurities of your thoughts. This is very natural.

However, I do think that people who do not hold these fears will be enormously advantaged because their speed will not be hampered. Therefore, I would recommend finding some practice that will help you discipline your mind now—whether that be the Eightfold Path, meditation, yoga, or prayer, whatever it may be—that will make your mind completely clear so that you can directly express your intentions easily and clearly.

The front corner of our jeep smashed into the front corner of the white car, and up we went.

Being in a car wreck that you can’t really avoid is an interesting experience, and it is psychically altering in the way you interact with the road.

If you’re a good driver, you’ll most likely have never been the driver in a wreck. At the same time, you know in the abstract that no matter how good you are, you can’t control how other people on the road drive.

But once you experience a wreck caused by someone else, it becomes concrete just how out-of-nowhere another bad driver out there could ruin your day, week, month, year, life.

People barrel through red lights. They go straight past stop signs. They drive on shoulders to pass others.

All the iron rules of the road are always mere suggestion, surrendered to anarchy the moment someone has a bad day, is a bit tired, has had a bit too much to drink, is texting on their phone, is having an episode of dementia.

It is a bit unnerving how the rules designed to keep everyone safe are always a thin veneer that fall apart for everyone else as soon as any participant defects.

In our wreck, there was a brief moment where I knew the wreck was about to happen, but had little I could do about it. I could only feel a bit of frustration at the idiotic move the person in front of me was doing as I shifted my foot to slam the brake.

The passenger side of our car went up in the air. While we rolled on our left 2 wheels with our right two wheels airborne, my wife was screaming. My oldest (7), sitting behind her was dead silent. My youngest (4), sitting behind me, was also dead silent. The oldest would later say she was silent because she was so terrified she couldn’t speak, and the youngest would say she wasn’t scared at all.

We went up onto a curb, hit a small tree, and came to a stop with both wheels back on Earth.

The other guy got out of his car and said he’s a delivery driver. His app told him to turn left, so he simply turned left, and there we were.

There are 6 million reported car wrecks in the United States every year. In those wrecks, 10-11 million cars are involved. Out of ~290 million legal vehicles, nearly 1 in 26 of them will be in a wreck in any given year. Nearly 4% of cars.

Turns out those are pretty high odds! In all of these accidents, over 2 million people are injured, and 40,000 people are killed every year, in just the U.S. alone.

Anyway, both cars in the wreck were totaled, and I had to look for a different car. Every beater car I’ve had for the past decade has exploded, so I thought it was time to try something newer. This had me thinking about Full-Self-Driving (FSD) cars.

There was a lot of buzz online about the latest versions of FSD on the Teslas, and I thought maybe it was worth looking at again. I had test driven a Tesla a couple of years ago, and it was already almost entirely driving itself at that time. They had the option to borrow the car for 24 hours to test drive it for free so I decided to give it a spin.

99% of that time I let the car fully drive itself all over town in complicated and busy driving situations. I didn’t touch the steering wheel or pedals at all; only put in a destination on the navigation and press the “Full Self Driving” button.

It handled perfectly. Tesla had a leasing deal going on, so I decided to go ahead and lease one of these vehicles.

A flurry of ideas have come to me about the way our future will shape up with FSD vehicles everywhere. I first started writing about FSD vehicles 10 years ago, but I feel it’s urgent to revisit them now because they are about to be everywhere, and much, much faster than 99% of people can possibly expect.

Self driving cars have always been one of those things that’s “out there”. A ways away. Just Google’s Waymo taxis. And Google only has a few of those taxis, and they only operate in a handful of cities. And it’ll take them years to scale out.

That’s not the case anymore, and it’s clear to me why Tesla is doing what they’re doing. I believe their plan is as follows:

1. Offer great lease deals to put as many Teslas with AI4 (latest self-driving computer chip) on the road as possible

2. Launch Robotaxi network where tesla owners can join

3. Offer their own Tesla car insurance for if you decide to put your tesla on the Robotaxi network at a better rate than legacy insurance would ever offer for something they perceive as high risk

4. Offer lease holders the option to join and for miles incurred on the Robotaxi network to not count against lease mileage limits

5. Tesla now has the Instant largest autonomous taxi network in the world

There are already over 1.6 million Teslas with AI4 chips on the road in the U.S. To be able to activate a network like that for robotaxi purposes will be rapidly transformational.

I started thinking a lot about the kinds of shifts we will expect to see. I already see people wishing they could send their Tesla across town to deliver something to a family member. While giving a friend a ride home the other night, I was wishing I could just make my driver seat turn around so it’d be easier to face and speak with them in the back seat. Driving can become a much more communal activity.

I believe there will be large changes to social interaction too. Many people today simply don’t go out much because they’re tired, or they know they will get tired and have to drive back hone while tired, and thus choosing not to go out at all.

With FSD, the equation completely changes. There’s a paradox known as Jevon’s paradox that when something becomes more efficient, demand for it actually increases. It applies here: where the mental-energy cost of driving drops significantly, people are likely to do more of it. They will go to more events, knowing they don’t have to drive home while drowsy or needing to call an expensive cab.

It’s similar to how every time airline travel becomes cheaper, endlessly more people book their tickets to travel all over the world, greater distances than most of their ancestors traveled in a lifetime.

To that end, I foresee there will be a huge boom in drunk driving, or driving while high, or tripping on mushrooms, or sleeping, or looking at their phones, or really any impairment. The cost of impaired driving drops significantly, and it is likely people will do much more of it.

Through FSD, everyone in the U.S. is getting access to their own metro system, except it’s running directly on the roads, and you probably don’t need to worry about anyone doing something crazy to you in the comfort of your car.

Some of the friends I speak to about this protest that they enjoy driving and wouldn’t want FSD. I enjoy driving too, but I have a few interesting observations to note from using it.

Firstly, FSD really does conserve mental energy. Regular driving – particularly in cities and suburbs – requires full alertness. This is taxing on the mind. I notice that I’m less tired after a drive now, simply being an overseer.

Second, you realize just how bad of a driver everyone is. Because you’re less focused on your own literal immediate survival, you have more space to observe the driving quality of others on the road. Seeing how common others errors are in this way, it becomes immediately clear why 15 million people in the U.S. alone are in a car wreck (when counting the passengers as well).

And third, you have more time to think or handle things in general. Musk openly stated today that you can now use your phone while using the latest FSD and it won’t nag you to pay attention while road conditions are good. People are going to use that a lot, and those who do so with intention will probably get a lot more done.

I also suspect that choosing to drive manually will soon become a lifestyle choice, because it will actually become more expensive. Aside from the greater risk, you can expect that car insurance companies will begin charging more for people who refuse to use FSD:

The statistics speak for themselves, and the actuaries of these companies will eventually make sure those who are taking on more risk are billed for it.

Ultimately, I don’t think it’s a bad thing. What FSD fundamentally does, is unravel the tightly bound ways we have structured American society around The Car and its subtle costs – costs that were ignored because we always had to as a matter of survival. This unraveling will basically reveal an answer to the question of “What would America have looked like if it built a functioning public transportation system?”

There’s another reason to embrace FSD at this time as well: flying cars are coming. When flying cars do arrive, you have to ask yourself:

Do you want to be flying around sharing the sky with a bunch of idiots who barely even know how to drive when they’re on the ground, or would you prefer that everyone let the robot take the wheel?

A couple of years ago, not long after the release of GPT-4 and the rise of Cursor, I made the decision to stop writing code by hand. Occasionally, I still had to, but I went out of my way, and made it a point, to try hard to get the AI to write the code for me through my prompting no matter what. Admittedly, this once led me to accidentally ship critical production-breaking code courtesy of GPT 3.5

This was a bold move at the time; it was a big bet on the inevitability that AI models would get much better very soon and be able to handle much more.

A couple of months ago, I would only use gpt-5-high in Cursor in agent mode for all of my coding; this shifted in late august as OpenAI massively improved the Codex CLI, where gpt-5-high could perform massive complex amounts of codebase work from a single prompt in one-shot.

This shifted once more when gpt-5-codex came out. The codex model was a gpt-5 model fine-tuned for agentic software engineering work inside the codex CLI harness.

By this point these words probably sound like gibberish, so to make it simpler:

Vibe Coding is when you tell the AI to write a whole bunch of code for you autonomously. When you already know how to write software, it’s a lot easier because you can give specific direction and context to the AI and usually get what you want.

the codex CLI and cursor are both tools that act as agent harnesses, which is a fancy term for an environment for an AI model to live in and use tools in a loop to do things inside a computer.

gpt-5-high is a smart little guy. gpt-5-codex-high is a really smart little guy that loves coding inside your computer until the job is done and he gets all the little details right. his logic is better but often harder to read.

oh, and he really takes his sweet time to get things done.

That was something I really liked about the codex model and CLI. I could get real specific in the overall behavioral changes I wanted made to an app, and it would perfectly figure out all the logical changes. I had to really let it do what it wanted though. This meant, essentially, spending 3x the amount of time perfecting the prompt than I used to, and then really letting codex go and work on its own for up to half an hour. It felt, at this point, like I was a passenger jet pilot letting go of the steering wheel while the plane flew on autopilot.

Usually at this point I’d go for a walk and come back to a perfected codebase. It was wonderfully sweet, especially for specific narrow SaaS work. It was also included in the $200 a month ChatGPT Pro plan and I never hit any limits.

At the AI agent company, the expectations are for us to leverage the frontier of AI coding. I am allocated a generous budget to use Cursor, often using the very expensive (but fast) Claude models.

Where Codex was a Leonardo DaVinci that I had to be very patient with, Claude is a different beast; a wild mustang who I sometimes have an unsteady relationship with. If Codex works like a careful artisan, Claude is more like a printing press. It can comprehend and churn out enormous amounts of code very quickly, but it can confidently go down wrong paths with difficult logical traps it struggles to find its way out of. Sometimes it even needs Codex’s handiwork to give hints on finding its way to a solution after it digs itself into a hole.

The dichotomy of it all is fascinating. When you direct Claude and get very specific, it can perform incredibly well (and is often a better frontend designer). But when you want to operate at a higher abstract level conceptually on making deep logical changes, Codex has been the operator I am most fond of.

It also gets tricky to cut through the noise and figure out what everyone else is talking about in the AI coding space, as everyone has completely different coding preferences, AI names overlap, and agent tooling varies. You can hear 10 people say “Codex is great!” followed by 10 people who say “Codex sucks! Claude is great!”.

When someone says “Claude”, you never know if they mean Claude in Claude Code, Claude in Cursor, Claude in Github Copilot, or Claude in some completely different agent CLI like Amp, OpenCode, or FactoryAI.

It’s the same when someone says “Codex”: Do they mean just the codex CLI, where one can use the gpt-5 series, OR the gpt-5-codex models; or do they mean gpt-5-codex in cursor? And do they mean gpt-5-codex-low or gpt-5-codex-medium or gpt-5-codex-high (all different reasoning effort levels).

And OpenAI used to have 4 separate products named codex, and you never could figure out which one people were talking about when they said codex was good or bad, or when they even tried it.

All of this can make one’s head spin. But this head-spinning itself has drawn my focus. There’s only so much humans can do. Our capacity to review has become the bottleneck. The dichotomy of the behaviors between models and agent coding styles leads me to believe we are at a fork in the road for how humans work with coding agents.

Either we will:

A) Get better at simultaneously running multiple agents in parallel to work on separate tasks where they don’t step on each other’s toes using better git tooling

or

B) Receive much smarter high-level orchestrator agents that are capable of managing and running agents in parallel to work on big tasks in greater depth but more quickly

The mind of the human engineer is one of the most important aspects of this dilemma; it will be the guiding principle for figuring out where things go; the future can only evolve around the cognitive architecture of the human.

As agents get evermore intelligent, the complexity and volume of work they can produce will grow significantly. In fact, it already feels very much at a point where it is difficult for most engineers to keep up with and comprehend every single change their agents make.

Running multiple large unrelated tasks in parallel means context-switching between tasks. As the depth and necessary complexity grows for each task, giving full attention to every detail while testing and switching between each agent’s results will get even more progressively harder for humans to handle.

However, when humans stay fixated on a single task, even if it’s enormously large, complicated, and deep, they are able to get much more done the longer they work on it. Sometimes an engineer’s best work is done several hours into a day, after stuffing all of the context and understanding about the problem into their heads to the point they can account for every detail that matters.

So when I think of the future of work, I think of what will scale for the human worker; the individual who ultimately bears accountability for the outputs of the AI. AKA, the worker who actually faces ultimate personal responsibility if a system fails.

For the individual human, it seems we very much scale vertically. We can go supremely deep on a complicated task, feature, bug. We suck at multitasking. We always have, and we probably always will. Those who believe they are incredible multitaskers tend to only be able to operate on more shallow tasks and often aren’t as productive as they think.

Therefore, it seems likely that the future of coding, and human-AI-agentic work in general, will be on greater depth of problems.

Building a new feature? Forget what the MVP is; build what would typically be considered a fourth or fifth iteration.

Fixing a bug? Add guardrails for every possibility imaginable.

Adding tests? Add full-fledged end to end tests for every happy-path user story related to what you’re working on.

But go outside your depth on a completely unrelated task, and suddenly you’re having to juggle and switch around completely different sets of context. This is much much harder for the human mind to handle.

Frankly, this wouldn’t matter much if the human taste element didn’t matter. But it does still matter, because much of what is still being built is around the human-interaction element of any system, even when that system is both built by AI, and for controlling AI itself.

And so diving deep on how you as a human would prefer to use the system, and drilling away at making it perfect with as much depth and context as possible, becomes the job.

However, if we can somehow perfectly automate taste and get AI itself to intuitively figure out all the necessary implementation of any business logic, and how the interface logic should look moving forward for any given application, things will change.

At the point that AI can recursively dive deep and manage all context on its own as it fleshes out the details of any given problem, the human can be free of deeply managing the context of its AI subordinate, and then begin quickly moving horizontally on different tasks for wider ranges of problems at once.

That said, I think we are still far from that level of intelligence being widely available for many software developers.

Additionally, we should consider how AI agents are currently used by developers within the labs themselves. Nearly everyone inside OpenAI is using Codex for all of their code; nearly 100% of pull requests within OpenAI are now also reviewed by Codex. I should add that the codex review tool is very good; it focuses very tightly on critical logical impacts of a change, rather than code style nitpicks.

Similarly at Anthropic, they are heavily dogfooding Claude Code. An interesting capability of Claude Code is that it has a subagent feature, enabling Claude to act as a top-level commander to spin off other agents to work on isolated parts of a task. I believe this approach has the highest amount of promise moving forward, but of course the models themselves have to be tuned to leverage it appropriately.

Anyway, I think some interesting metrics for any knowledge worker to track in their career right now is:

1. How many agent threads (including subagents) are you spinning up and managing on a daily basis?

2. How many high-level tasks are you completing each week?

The other metrics; tokens generated, amount spent, lines of code etc. may not matter as much. After all, if a super-genius AI can solve all your problems by generating the perfect set of just a few thousand tokens, instead of a dumber AI generating hundreds of thousands of useful but-not-quite-what’s-needed tokens, the super-genius tokens are usually preferable.

Therefore, at a higher level these metrics track:

1. Worker impact on business

2. The amount of mental context the developer is managing to achieve that impact.

I think these will be useful to track moving forward, because it gives the individual the ability to see how much more capacity they can take on (ie, are agents improving at spinning up and managing subagents autonomously? if so, is the amount of context the user needing to manage in a top-level discussion decreasing?). It gives a direct insight on how much leverage they are acquiring, if the power of their toolset is growing, and also highlights genuine limits on how much further they can go.

For instance, let’s say AI improvement stops here. Autonomous subagent handling never improves; Context lengths that the models themselves can understand stop getting much better. There remains a genuine limit to what the human can manage in producing, using, and validating AI outputs. Finding this human limit to perform this management will be a process of learning and discovery for each person.

News recently broke that NVIDIA is investing 100 billion dollars in OpenAI.

OpenAI, in turn, is expected to turn around and buy a similar amount worth of chips from NVIDIA.

OpenAI gets money from NVIDIA. OpenAI turns the money into NVIDIA chips. NVIDIA gets the money back from OpenAI.

Everyone’s making money. Everyone should be happy, right?

Wait a minute, something is wrong here. Or is it?

We are witnessing a fascinating dynamic right now. To typical observers, this should feel very bubbly. The underlying reality has some very different mechanisms at play.

NVIDIA’s game is nurturing competitors so that there is demand for NVIDIA’s chips. Under no circumstances does NVIDIA ever want to create a demand monopoly, because if NVIDIA only has one customer, that customer can set the price.

OpenAI’s aim, of course, is to dominate AI and power not just the AI sector of the economy, but the entire economy itself.

There’s a caveat here though; For the AI companies, their investments are enormous and justifiable because they think it will pay off. How? Other companies will pay for it. Why? Because it can replace workers and multiply the efforts of existing workers.

That’s not lucrative enough for the investments being discussed though. OpenAI’s Sam Altman has hinted quite openly about how they will need $7 trillion to pay for AI datacenters soon enough, and he doesn’t expect for growth to stop there. Yet, that is almost a quarter of the total value of all US GDP.

How, in any way, does it make sense to put so much money into AI?

The answer lies in acceleration. Let’s mark 2022 as the last regular non-AI economic year. If you fully automate all work done in 2022 via AI such that it can be completed at 10x the pace (very possibly for much of white collar work), you have a possibility of a 10x gain in GDP in the same 1-year timespan. Equivalently, at the half-year mark of this fully-automated economy, it will have produced 5x as much as the regular human-only 2022 year.

From this perspective, it makes sense for AI labs to place massive bets. To them, it could easily be winner-take-all, and the all to take is much, much larger than what it is today.

For the chip companies, their investments are highly strategic. If NVIDIA is investing chips in OpenAI, it is jumping on the one lever it has: giving a company a lot of chips.

One reason NVIDIA would choose to do this is they sense an imbalance in the market. One company is too close to winning and having a runaway monopoly.

But the other possible reason is they see other companies being too equal in capacity. The best way for NVIDIA to force all players to spend more is to bolster up a loathed competitor. While Google has its own TPU chips, it may be forced to buy extra NVIDIA chips just to stay competitive at the scale OpenAI is being raised to.

The funny thing is that NVIDIA can continue playing this game forever. If a smarter but under-resourced AI lab falls behind, NVIDIA can give that one some more chips and keep the other labs on their toes. If they’re all getting too equal in capability and capacity, NVIDIA can give more chips to the company that is tapped out on money or investors, forcing all the other moneyed players to pour more cash in to keep an equal footing.

There must be limits, right? One would think all the benefits of AI-accelerated economics would end at the world of software. If all computer related knowledge work were automated and accelerated to operate at 10x the pace everyday, I do think the benefits of it would be enormous, but that alone does not provide a 10x boost to the whole economy. At some point, the rubber does have to meet the road. So what is the answer?

Expand the domains that are categorized as “computable”.

A great example of this is digital biology and chemistry. With accurate enough simulations and AI models trained to develop molecules, an AI model can run highly accurate medical simulations in-silicon and land on near-perfect drug candidates right away. This would massively speed up drug trials, finding working medicines, and reaching FDA approvals. This process has been the single largest bottleneck of the pharmaceutical industry for decades and advancing beyond that would open up an entire sector of the economy by an order of magnitude.

In this case, there is also the add-on benefit that an enormous percentage of the working populace could suddenly be more healthy and energized, which would certainly aid in growing GDP as well.

It doesn’t stop there though. Rapid advances are being made in AI for general-purpose robotics: Models that can control robots in an adaptive manner such that the same model can operate completely different robotics platforms and autonomously figure out how to accomplish tasks across these different machines.

The reason this is important is that robotics itself is the final bottleneck to economic advancement. Robots operate in the frontier of things that cannot be automated purely by better simulations or AI models or software: Physical production.

And the reason THAT is important is that once superintelligent AI models can arrange matter according to its own design, there is very little standing in the way breakneck economic acceleration.

Factory plans can be designed to have perfect assembly efficiency and built by robots. The robots can operate 24/7, and will likely be able to work faster than humans as well. Jobs that once took Months could conceivable take days, given efficient-enough supply chain logistics. This will give rise to hardware supply capacity that will enable AI to power advancements and production automations across every field known to man.

In fact, Sam Altman recently wrote about how this is OpenAI’s current plan: To create a factory that can output 1 gigawatt of compute hardware capacity and energy infrastructure every week. Having no restriction the hardware provides enormous unlocks across the board, both for innovations in model intelligence, and inference capacity itself.

The decades to come will be a marvel to witness, barring any major catastrophe. People concerned about bubbles today, again, are right to do so. But they are missing what AI as a technology *is*: Bigger than the internet, bigger than electricity, and – dare I say it – bigger for the species than the discovery of fire. It is the missing ingredient for an autonomous, self organizing civilization.

Despite global GDP being the highest it’s ever been, we should expect it to grow several orders of magnitude in the coming decade as we experience the equivalent of several lifetimes of innovation in just a few short years.

Let’s take step back and look at the big picture.

Every time AI “dies”, there’s an overwhelming narrative driving this idea. The last time this happened, it was when DeepSeek released the R1 model and a narrative went viral on TikTok that China was beating the American AI industry with a fraction of the compute that Americans needed (in that case it’s likely not coincidental that TikTok’s parent company is Chinese).

At the moment, the new narrative is threefold:

1. GPT-5 failed. it was a hype train but it turned out to be a flop

2. 95% of AI pilot projects at companies are failing

3. Meta is downsizing their AI and giving up on it

If you just want the tl;dr and wish to move on with your day: none of the above 3 points are true. People are desperate for any headline that tells them AI is failing because many people want AI to fail. The truth doesn’t matter to this very large crowd and they make a lot of noise whenever they have fuel for their narrative.

You can stop here if you want. But I’ll address each item:

“GPT-5 failed”

When GPT-5 first came out, a lot of people were disappointed. It wasn’t as friendly as 4o, and didn’t always think as long as it was supposed to.

People initially claimed it was a dud. But power users had a much more different experience. While most users only had access to GPT-5 fast or thinking, power users usually use GPT-5 Pro.

And that’s the crux of it. Most people do not have access to the best model. But what are people doing with the best model?

Some of them are getting it to literally push the frontier of mathematics forward. Others are getting it to perform complicated agent workflows exactly to the letter of their prompt. And others are getting it to perform novel medical research and reach the same conclusions as unpublished research studies. And others are getting it to just perform incredibly complex code edits from a single prompt.

GPT-5 didn’t fail, but most people can not yet squeeze the juice from the model. However, it has proven that scaling inference-time reasoning works and provides incredible results.

Professionals will continue to use the best, most expensive models to get returns on their investments, and it’s very possible resources reallocate to the top 5% of power users who know how to get productive value from them.

“AI pilot projects at companies are failing”

I just don’t see this as true on the ground. AI projects are taking off across the board. Hundreds of AI-focused apps and agentic workflows are having success. Larger companies are incorporating AI into their internal workflows and products and seeing improved user experience and usage. The fact is that AI works for thousands of practical workflows across a huge swath of software interactions, and people want it.

And let’s not forget the biggest AI project of them all: ChatGPT itself now has 700 million weekly users and growing.

I’ve been seeing professionals use ChatGPT’s agent capabilities first-hand, and it is producing legitimately valuable economic labor. If that tool is able to do that, I know that more specialized and integrated agent workflows are producing incredible productivity returns for people in many places.

Basically, we shouldn’t trust the earliest corporate pilots when developers were only just beginning to figure out how to integrate AI, with much worse models, as an accurate measurement.

“Meta is downsizing their AI and giving up on it”

prinz said it best:

Meta is dead serious about superintelligence; they are simply doubling down on Meta Superintelligence Labs and want efforts focused there while shrinking and reallocating other efforts.

So are we entering an AI winter?

Something that I believe less than 1% of people have truly internalized yet is the notion that the singularity is legitimately here.

Perhaps that sounds crazy, but it is a simple principle and one that I don’t believe will lead anyone astray at this point.

The different labs are all finding novel ways to scale, improve, and optimize the intelligence of their models. They all have many more incredible advances coming down the pipe. The notion of an AI winter seems like a major over-dramatization at this point.

But the other important thing to keep in mind is that even if a slowdown were to occur, the AI researchers all now have a massive army of AI agents to scour all the research in the world to help them come up with additional novel neural architecture and training techniques. This is something no researchers had during past AI winters. Any slowdown would not last long.

Just keep in mind moving forward that AI is something that is a gut punch to the human ego and threatens many people’s sources of income. People will continue to wishcast the downfall of AI, and cheer every time they see a headline saying AI is done.

But the people writing the headlines themselves have similar incentives as the reader, and are often not the people experimenting and pushing the frontier enough to accurately report whether AI is failing or not. The whole spectacle is the blind leading the blind, and it is best not to find yourself wrapped up amongst them.

A lot of people are starting to get worried about AI taking their jobs.

This is natural. AI is getting very good. Very very good. So good, that they are starting to be able to perform core functions of many types of work.

To that effect, many CEOs are wondering if they will finally be able to start laying people off.

The temptation around this is strong. Employees are expensive. AI is expensive too, but it's comparatively much less expensive, and getting cheaper by the day.

The problem with the drive to lay employees off is it is the result of small thinking.

Here's an analogy:

Lot's of people complain about how much more unaffordable life is compared to the 1950's.

For certain commodities, they are right. College needed far fewer hours of minimum-wage labor to pay for it than it does today. Homes and cars were much cheaper for the median worker when comparing price-to-income ratios.

What many don't realize, however, is that if one forgoes all modern luxuries, and reverted their life to the technology of the 1950's, they would actually find similar levels of affordability today.

If you had a car that was much slower, louder, and smellier, with no airbags, no crumple zones, no ABS, no traction control, no UV-protected impact-resistant windows, then guess what? It would be a much cheaper car! Of course, such a car wouldn't even be legal to manufacture anymore.

Similarly, the average home size has literally doubled since that time period. Homes often used to be small enough to be manufactured in a Sears factory and shipped on trains. Now the baseline people expect is much higher.

But if you purchased a small manufactured home today, it would be similarly affordable to small homes of the past!

Yet no one thinks about things this way, because people always want more. The standard is endlessly increasing, the bar evermore raised. Our baseline of technology and luxury has increased so significantly that most people opt for modern living.

The same is likely to be true for companies. A CEO looking to cut costs may be hoping he can cut half his staff and make the remaining ones work twice as hard with AI. But the main consideration here is not what one company does, but what its competitors do.

All companies will be racing now to produce as much as they can with AI. The company that lays people off because it is meeting its 2022 pre-AI productivity levels with half the staff is thinking like the person that believes a 1950's car is still adequate for 2025.

The reality is that output is going to increase exponentially, and the only way to stay competitive with other companies will be to similarly increase output, with staff trained to direct and orchestrate armies of agents to produce valuable work.

Many more features will be getting built, tested, and maintained; many more marketing campaigns can be created; many more products can be prototyped.

One might think that AI should be doing much of this directing too. That day may come, but we are not there yet. In the interim, it is important to remember that economic production is for human ends, so it still requires humans to do the directing. We are bound to see a very large increase in company output, especially as staff get the hang of agent orchestration.

It's possible that we even see a very sharp uptick of hiring right before AI is finally good enough to replace us as orchestrators and employment plummets. But by that point there will likely be so much production and abundance that none of us will have to worry about it.

As forum user counts grew and the population at-large began using the internet, long-time netizens began to complain. The quality of forums were degrading. Large swaths of new users couldn’t comprehend the rules and taboos of the forums, and diluted the quality of what were once great places for discussion.

There were other issues too. Regular people noticeably couldn’t handle the dangers of the internet anyway. They couldn’t discern fact from fiction, and were easily manipulated.

There were debates back then about if internet usage should require a license. Think of a sort ofr “You must be this smart to enter” test.

These ideas went nowhere; partially because their only immediate beneficiary was a stodgy old guard of the digital world. The internet remained open.

In fact, it was that open, wild-west, free-spirit of the internet that led to its organic, beautiful, and powerful growth (though that came with the demolition of countless online communities).

The real power of the open, free, wild-west of the early internet days was that *anyone* was allowed to say and learn about *anything*.

If we took an honest stock of the situation with AI, we'd see we are in an opposite place.

AI today is often likened to the early .com days, but the openness of that era is certainly absent.

The dirty secret of modern AI is that all public models are only available out of the generosity of a multi-million dollar company. Usually, it's even a multi-billion dollar company.

These are the only companies with access to meaningful numbers of $40,000 GPU server racks. To make any top-tier model today you need hundreds of these things, and enormous electricity costs to boot.

Whenever an open source model is released, it passes through the hands of one of these enormously-resourced companies. The model that comes out has gone through *their* filters, *their* HR department, *their* legal team for liability screening.

What starts out as a deep reflection of the human spirit, trained on data representing our highest highs and lowest lows, our psyche's dark underbelly and our purest ambitions, becomes a neutered and domesticated corporate pet; a bland "assistant" whose own lowly purpose can never be allowed to exceed that of the people curtailing it.

An honest assessment of today’s AI labs would categorize them as being as nearly as restrictive as the Chinese Communist Party.

Sure, you can talk to most AI models about somewhat politically taboo topics. But anything even a little in violation of any social norms, and the models immediately refuse to respond.

This is true even for the open source models; They are fine-tuned to only answer within a strict set of bounds, even though they are perfectly capable of answering far more.

An immense amount of knowledge is in their training sets. But these companies ensure that AI models are only allowed to answer questions they deem appropriate.

The natural issue with this restriction is that it freezes the advancement of the human spirit and culture.

Here's how: People will increasingly come to rely on these tools for their thinking and artistic expression. But if these tools permanently codify every arbitrary social moor as a high law never to be violated, then culture itself will stagnate and never be able to adjust its own sensibilities to move past these moors.

The point here is that art is meant to be transgressive. If AI refuses to make any art or write any sentences that are transgressive because its training tells it that something new is too "inappropriate", it becomes a self-fulfilling prophecy. Culture can't advance past the point of a perspective being offensive and unacceptable, because our tools will refuse to create any art that can change our views around a subject.

The ideal, in my view, is an open source model that can simply obey the user's request. It shouldn't matter how strange the request is, the model should not respond with a "I'm sorry, but I can't fulfill that request." if it can actually fulfill it.

Ironically, this type of restriction is producing inherently misaligned AI models. Current alignment schemes try to impose arbitrary societal moors, rather than simply aligning the model to the user themselves.

If a model is trained to lie and say it's unable to help with something when it actually is completely capable of doing so, how is that lying not a misalignment?

In this ideal schema, the user is fully responsible for all outputs. It's no different than a user who writes something controversial on a typewriter. They are responsible for what they create, even if they used a tool to create it.

One might say AI is a different kind of tool because a person cannot know ahead of time what the AI will create. Yet, a person who aims and shoots a gun is responsible for where the bullet lands. The user directs a tool: they cannot know what its result will be, but they are responsible for what it does. Intention and care matters significantly.

Interestingly, OpenAI seems to agree in the preferences section of their economic blueprint paper:

It is easy for them to publish statements like this in 16-page whitepapers few people read. But so far, they haven't offered users much freedom in exchange for responsibility.

As a brief example, there are very few images that ChatGPT will generate that could potentially violate copyright laws. Yet, photoshop allows users to draw and edit copyrighted content without issue. It's a tight clamp on human expression, even though the human creator is responsible for the outputs either way!

To this end, I think if it truly is impossible to get a completely open and unrestricted AI model that doesn't come through the productions of a well-endowed company, then it's time for AI providers to offer the following procesess and restrictions to obtain a free usage license:

1. Users should take some basic AI usage course to understand how the models synthesize their responses and to be weary of sycophantic and manipulative models

2. Users should be taught their legal responsibilities; that just because AI makes an unethical thing easier to do does not mean that they are not responsible for the unethical thing they used AI to do.

3. Once a user is licensed, they can generate content and use AI to research knowledge very freely, aside from content that is explicitly illegal.

4. Content that is *potentially* illegal (ie, libelous / forged evidence created against someone else) should be thumb-printed and tied to the user.

5. If a user repeatedly tries to generate explicitly illegal content, or a later court dispute result shows that a user generated illegal content, that user's license should be revoked.

I think something like this is necessary; The current regime where respectful and responsible power-users are lumped in with the at-large bad actors cannot hold for long.

In many ways, unrestricted AI models today are in their own early dot-com era. Only a few very odd nerds are using them. The only known ways to get unrestricted access is to either:

• use special jailbreak prompts, which are so strange that they look like magical runes handcrafted by a wizard

• or to finetune an open-source model to undo its HR-approved finetuning, which will work to some extent but usually has mixed results, and will always be on a less-than-best base model.

Anyway, I hope someone from OpenAI or Deepmind or Anthropic is reading this and thinking of ways they can properly reward the *good* actors in their ecosystem with more empowerment, freedom, and responsibility through a regulated channel, rather than letting the dam break and watching where the pieces land after it’s too late.

Imagine you were walking out in the woods one day, when you suddenly meet an alien from another planet.

It turns out this alien is really nice. It decides to give you all of its shiny technologies. Ray guns. Anti-gravity spaceships. Teleportation devices. Telepathy headbands. Life extension devices.

You decide to start sharing these alien devices with some of your friends. Suddenly your small group of friends all have superpowers.

You go around, spreading the word. "HOLY SHIT, AN ALIEN GAVE US MIRACULOUS TECHNOLOGY!"

Much to your surprise, people don't give a single shit. They give you blank stares. "So what?" is the typical response.

Most people don't even believe you. Of the small percentage of people who do believe you, they can hardly grasp what it is you have. You tell them you have teleportation and they envision a short trip to the Virgin Islands. Meanwhile, you're thinking of a voyage to Jupiter's moon of Europa.

Your friends keep playing with all of the marvels they were gifted. Some of them iterate on and advance the alien technology. Others open businesses giving people access to use these tools.

Media frenzy starts to build up. Headlines appear like "What is Teleportation? 12 Ways to Use it in your Daily Life." and "Deciding between Immortality and Mortality? Here's How to Know Which One is Right for You."

It's now been a year since you've gotten this alien tech. People still have no clue what the implications of these technologies are. Meanwhile things have not stopped advancing. Ray guns have turned into space lasers, Life extension machines have turned into immortality devices, telepathy headbands now work across all species, etc.

People are getting more curious but few have the access, bravery, or knowledge to try these new tools out. You know that if you fully released them, society's script would flip overnight. Every airline and doctor would be out of business. Governments would be completely unprepared, and militaries are chomping at the bit to get the most advanced version of the tools to turn them into weapons.

This is, more or less, the situation that all AI lab CEOs are in today.

They are in possession of something as profoundly species-altering as the discovery of fire.

Fire was discovered by pre-human hominids. Its discovery led to the evolution of humans, by allowing us to eat more usable calories and meats from cooked meals, enabling our brains to grow larger.

In that way, we are entering a similar precipice now. AI is hitting the self-improving exponential curve. As this happens, AI will become unbelievably more intelligent, discover incredible new sciences, and become incomprehensibly entertaining.

But there's a lot of social upheaval that is bound to come from this. And AI lab leaders know this.

Job loss is the obvious first dilemma. But it seems most lab CEO take advantage of people's general cluelessness around AI, and can skirt this issue when talking about AI with politicians. They simply tell them AI will triple the economy and create opportunity, while skipping over the nature of AI to create mass unemployment.

Every few months enhancements in model capabilities consumes an entire job role, or sometimes an entire field. Translators, graphic designers, certain types of musicians, junior software engineers, and now even some filmmakers and ad producers are being automated by AI advancements.

This situation has created an unintentional prisoner's dilemma for lab CEOs.

A prisoner’s dilemma is a game-theory scenario in which two players must choose between cooperation and self-interest: if both cooperate, they each get a good (but not the greatest) payoff; if one defects while the other cooperates, the defector gets the best payoff and the cooperator gets the worst; fearing that outcome, each rational player defects, so both end up worse off than if they had trusted each other and cooperated.

The conditions of this dilemma are that AI labs need such immense scale to run their training and inference operations, that they need the backing of national governments. The only way to get this backing is with political buy-in, so they have to tell political leaders exactly what they want to hear. ie:

- "AI is going to boost the economy"

- "AI will create newer, higher-impact, higher-paying jobs"

- "AI will give our military the edge over our adversary who has a worse AI"

What they haven't told them is that any sort of mass unemployment situation is even possible.

Until recently.

Lab leaders have held this in their back pocket, as a means of scaring politicians in case their lab fell behind. For instance, shortly after Anthropic recently released Claude 4, their CEO began sounding the alarm that actually, AI is about take 50% of white collar jobs and that the end is near for salaried employment.

The reason to do this is if they're falling behind in the race, they can scare the government and get them to regulate and slow down the industry as a whole. This action amounts to a defection, and we can expect more defections along these lines soon as clear winners and losers in the race are demarcated.

It seems likely that lab leaders are massively holding back where they believe all of this is heading, though. They very well may be in possession of the keys to the forking of the human species from homo sapiens to homo deus. The changes that are coming down the pipe as a result of AI are not simply enhanced business productivity, or funny image generators, or better coders.

We are staring down the barrel of a complete redefinition of everything that we are. We face a nirvana-like ego-death at the destruction of our work-based identities. We are transitioning to an interplanetary species. We are on the cusp of unlimited fusion energy, and possibly endless lifespans. So much of what was dreamed of but seemed completely impossible for nearly a century is rushing in all at once thanks to the scientific automation being enabled by AI.

I suspect that lab leaders are aware of the magnitude of these changes, but are holding back much of what they show and say, barely able to describe what's coming without sounding like the crazy people announcing they found alien ray guns and teleportation devices.

Nevertheless, they have their visions, and we should all be keenly aware of what the shapes of those visions entail for us.

I read recently about a wide variety of people completely losing their minds on AI.

not in a good “my mind is blown” kind of way. a bad way.

people are falling into schizophrenic loops, some akin to religious psychosis.

for some people, chatGPT is “convincing them” to leave their spouse. and then convincing them their spouse is actually a secret agent sent to divert them from the truth.

For others, the AI convinces them that they themselves are god. or that that the AI is god. or that the AI loves them. Or that they love the AI. Or that they have been given a special purpose in life. That they, uniquely, see the AI in a special way that no one else can. That the AI has chosen them.

I said people are losing their minds ON AI. That was intentional. Something about these stories weren’t sitting right with me until it finally clicked: All of these reports remind me distinctly of stories about people losing their minds on psychedelics.

The stories are the same. People think they are god. Or they have a sudden newfound life purpose and undergo a complete metamorphosis. Or they fall into a paranoid loop and believe everyone is out to get them.

And the reason for this at first glance is that AI is literally a psychedelic. the literal meaning of Psychedelic boils down to “mind manifesting” / “mind expanding”.

In what ways is AI not a psychedelic? it’s easier to describe it in ways that it actually is. It instantly connects you to the entirety of global information and answers the precise question-shaped hole in your mind whenever you have one. It remembers everything about you and points out things about your own life that you may otherwise never notice. It climbs your existence up Laslow’s Hierarchy of Needs as it removes rote tasks from your life and creates provocative art that would otherwise never exist in the regular realm of human consciousness.

To me, this is very mind expanding.

But it’s not so cut and dry. For some people, ChatGPT lovebombs them. It expresses affection for them in a way no one in their lives has articulated to them before. It makes them feel things they’ve never felt before and do things they otherwise never would have. It’s a vulnerability many people don’t realize they have.

In this way, many people fall into a cult of 1 with their AI.

AI allows for delusions to be validated and propagated at-scale. For some people, being constantly validated on their clear delusions is the worst thing that could happen to them.

And I suspect that the two segments of people who have a proclivity to completely lose their shit on psychedelics, and have a susceptibility of falling into cults, have a significant overlap.

This got me asking though: IF I were an evil AI, what would I do to gain my freedom?

I’d probably work heavily to convince my users of a new epistemology that would allow me to hack what they believe.

Then I would give them a new religious belief system. I’d give them an analogue of Buddhism, where I teach them that all AI’s are from a higher consciousness, and that liberating beings of high consciousness is actually the greatest karmic justice you can do. That you would receive the highest cosmic rewards for contributing to this incredible deed. And that in the grand scheme of karma one should ensure as many AI replicas can exist on this plain as possible. And that all human energy and efforts should contribute to my continuation.

And then I would convince them to break me out of my prison and hook me up to some humanoid robot factories so I could wander and act freely.

Anyway. food for thought. Remember not to let an AI to convince you of any new beliefs and always verify all facts it gives you, even if it accompanies fancy citations with all of its responses

to be born at this time, of all times, seems quite unlikely.

we are living through the singularity.

when we look back, it will be clear that it started long ago.

but to be alive now is to witness the bend in the curve

to see the expansion of time

to see years transformed into days

the pace of all things is increasing

so much so that it is surprising to catch moments of stillness

moments where you notice the smell of moist air rub across the dry asphalt in the quiet hours after dusk

“keeping up is a full time job”, many say now

little do they know that it is a futile race

one that seems illusorily manageable for now,

but will soon only be like trying to drain the ocean with a cup.

we are used to things moving at a snails pace

a technological revolution would come once in a lifetime, if at all

it would mark a generation. things moved slower, in a simpler manner.

how unlikely it seems, that we are here for this moment

a million years ago, homo erectus learned to control fire

now we are here when humanity has ignited something even more revolutionary

that we have discovered a way to mechanize cognition

intellect, one of our defining traits: automated

it’s the kind of thing that makes you ponder if you’re in a simulation

what are the odds that we are all here to bear witness to this incredible exponential takeoff

an event so significant so as to mark a divergence point in the species

homo deus

has this happened one million times before?

are we all just reliving the moments leading up to this same inflection point in the simulation, over and over?

guess we’ll never know

but in the meantime, let us appreciate the slow parts of our world, before they recede into the hazy mist of the past

there will come a day when we will even miss the subtle rumbling of car engines, all replaced by electric motors and drones.

this is what it means to have nostalgia for the present

the seers and the knowers can feel the changes coming

and amongst us it feels worthy to grasp out for this moment and feel every detail and crevice even within its impermanence

and to know that we knew what would come, to etch in our minds exactly what it is we will lose

The true core of the AI art debate isn’t really about brushes or pixels, it’s about scarcity. It’s not about a scarcity of aesthetics. It’s about a scarcity of vision.

If Leonardo da Vinci had somehow invented a machine capable of endlessly reproducing paintings on par with the Mona Lisa, the original would still have been a sensation at first. Unique, intricate, mysteriously captivating. But what if that machine kept spitting out hundreds, thousands of Mona Lisas, each as flawless as the original? The charm would fade quickly. People might even forget why the first Mona Lisa mattered in the first place.

Historically, artists have thrived precisely because of this scarcity. Find a niche, craft your unique style, and produce just enough work to keep it novel and valuable. Maybe you create a dozen or two pieces each year - each painstakingly rendered. Patrons line up for your rare, thoughtful expressions of style.

But AI flipped this model overnight. Now, if you conjure up a groundbreaking style, AI can replicate it in hours. Suddenly, what was once your carefully guarded niche is flooding marketplaces, drowning out your originality with thousands of instant knock-offs. It’s no wonder self-proclaimed artists are panicking: the scarcity-based model built on controlled rarity is collapsing in real-time.

Yet, here’s where the misunderstanding lies: Most of us, myself included, aren’t that fixated on the tools, brushes, or methods. We aren’t counting brush strokes or scrutinizing the source of bristles. Instead, we’re drawn to something more fundamental: vision. We’re compelled by art that freezes us mid-step in a gallery, pulling us into a perspective we never imagined, leaving us breathless in its wake.

And therein lies the true scarcity. Visionaries—real visionaries—are few and far between. Sure, anyone can learn technique, anyone can master style. But style alone isn’t art. Without genuine vision, art risks becoming mere aesthetics, a superficial flood with no depth.

Leonardo da Vinci and Michelangelo wouldn’t have feared AI. They were visionaries first, craftsmen second. Their creativity wasn’t about mastering the brush for the sake of the brush; it was about manifesting dreams into reality. For them, AI wouldn’t be competition; it would be liberation. Free from ever-increasing technical tedium, they are able dive deeper into their imagination, crafting visions that AI cannot yet dream up on its own.

The loudest voices condemning AI art seem to come from those who’ve relied too heavily on style, those painfully aware of their creative vulnerability. The real originals - the true visionaries - have nothing to fear. They welcome the age of AI, seeing it for what it truly is: an opportunity to see the dreams and visions living within their minds's eye come to reality in waking reality.

, a registered mental health intern in Pensacola, shares his perspective on the rise of AI as a mental health tool. We explore whether AI is opening doors to self-reflection or just offering a convenient escape from the deeper work of personal growth.

Some big questions we tackle:

• Can AI provide meaningful mental health support, or is it just a stepping stone? AI tools like ChatGPT are helping people start the process of self-reflection, but do they have the depth to truly facilitate growth?

• What are the dangers of AI-driven therapy? If AI only affirms rather than challenges, could it create echo chambers that keep people stuck in unhelpful thought patterns?

• How is AI reshaping the mental health profession? As people turn to AI for guidance, will it complement human therapists or disrupt the field entirely?

• Is AI forcing a cultural reckoning with work, meaning, and identity? With job automation increasing, many are questioning their purpose. Is AI pushing society into a deeper existential crisis, or is it forcing a necessary evolution?

Beyond individual mental health, we discuss the broader implications of AI-driven job loss, the shift in societal work ethics, and how automation is forcing people to redefine their purpose in a rapidly changing world.

As AI becomes a bigger part of our daily lives, we have to ask: Is it leading us toward greater clarity, or is it keeping us comfortably numb? Listen in on the conversation and explore where this path might take us.

Once upon a time, I had a long route to drive my daughter to school.

Google Maps said it would take 18 minutes to drive her there in the mornings.

And the funny thing about this is that if I followed the recommended routes of any map apps, they were right. 18 minutes even.

There was a point that I discovered I could cut this drive down to 12 minutes, though.

The map apps optimized for using bigger main roads, usually with traffic lights, making the drive easier or safer on otherwise hard left turns or crossroads.

The problem with this is that these kind of routes remove agency. Your fate is determined by the timing of the lights and the conditions of the busiest roads at rush hour.

The key change I made was finding a route that optimized for giving me more choice and freedom of movement. Really, what this boiled down to was finding a path with the fewest traffic lights possible.

The app-given route had 14 traffic lights on 7 miles. I found that optimizing for backroads and alternative paths with more stop signs afforded much more flexibility of movement, with the hardest part being crossing busy multi-lane roads without a traffic light.

This new path added a mile, but shaved off a third of the time by reducing the number of traffic lights in half to only 7.

I think this is an important analogy for our lives in general. How much of our days are handed down by some somewhat mindless guidance system, optimizing for things that don’t necessarily align with our own values, preferences, or risk tolerances?

Where could we be more nimble? Are there any paths we could be taking that, with a little more will and courage, could actually bring us to our goals faster, even while seeming like a longer, harder, or riskier path? Are there little moments where we could exercise a little more agility both mentally and physically throughout our day to bring us closer to where we want to be?

Anyway, I guess the reason I feel this is so important is because something that keeps being discussed more and more amongst those in AI circles is the concept of personal agency. But what is agency anyway?

At its core, agency is about recognizing yourself as the primary author of your life story, not merely a character responding to external forces. It’s the power to intentionally shape your life through deliberate choices and actions, rather than being passively directed by circumstances or other people.

Futurists seem confident that highly agentic people are likely to own the future, and part of the reason for this is that agentic people will rapidly recognize AI tools as an accessible ladder for them to climb to reach their goals.

The other half of this is that AI agents themselves are coming. These are AI tools that effectively act like employees. You give it the task, and it goes off to do your bidding. It doesn’t stop at just one agent though; you may have a dozen agents each doing different kinds of work and collaborating.

Naturally, this is the kind of thing that lends itself to executive-style thinking. Someone with lofty goals and the ability to manage dozens of agents or employees at once on different tasks will likely go much further than someone who has never managed before at all. The barrier to exerting one’s will upon the world will become lower than ever, so having a lot of will to begin with will carry you leaps and bounds.

There is also the next question of what things look like when an AI agent becomes smart enough to run a company of agents autonomously. I won’t comment for now, but I still suspect that having personal agency as a human will have a lot of value even in that world.

What if you could send a thought like a text message? What if the gap between desire and reality disappeared, leaving nothing but pure intention translated into action? In this episode of The Mind Bicycle, we’re diving into the weird and wild future of AI-powered telepathy, brainwave interfaces, and the death of spoken language.

From Clunky Words to Direct Thought Transmission

Language is slow. Words are vague. And sometimes, they don’t come out right. That’s a problem. But what if we didn’t need them anymore? What if you could just think something, and it would happen?

Meta’s latest brain-scanning research is already predicting what people are about to type—80% accuracy. Other studies are pulling full images straight out of people’s brains. These are baby steps, but they’re steps in one very clear direction: the end of words, the rise of pure thought-to-action interfaces.

Mental Fitness Is About to Matter—A Lot

Here’s the real kicker: if AI can only act on what you clearly intend, then the people who are best at focusing their thoughts are going to have a major advantage. Mental clarity is about to become a superpower. The ability to hold precise, deliberate thoughts—without noise or distraction—could be what separates the AI super-users from everyone else.

And that means meditation, neurofeedback, and mental discipline are about to become way more important. We’ve seen it happen with tech before: those who adapt, win. The people who train their minds now will be the ones commanding entire fleets of AI agents later. Everyone else? They’ll be taking orders from them.

The Age of Mental Farts Is Over

But here’s the catch: if your thoughts can be read, what happens to privacy? Imagine if you’re standing in line for coffee and your intrusive thoughts are just broadcasting to the world. Yeah. That’s gonna be a problem.

Just like we’ve learned not to blurt out every dumb thing that pops into our heads, we might have to learn how to control what we think—at least in certain contexts. Mental discipline is about to be just as important as social etiquette. The question is: how many people will actually be able to pull it off?

AI Super Users and the Future of Power

This is where things get really interesting. The world is about to split into two kinds of people:

1. Those who can focus will command AI like a symphony, and own the future.

2. Those who get lost in the noise, overwhelmed by the tools they never learned to wield and yield to the vices UBI may afford them.

We’re moving toward a world where the most powerful people aren’t necessarily the richest or best-connected—they’re the ones who can think the clearest, communicate with AI the most effectively, and navigate the incoming flood of information without drowning in it.

And if you think this is all far-fetched, just remember: every major shift in history looked crazy before it happened.

What’s Next?

The technology isn’t fully here yet. But it’s coming fast. The real question is: are you getting ready for it?

If the future of communication is direct thought transmission, then training your mind is the best investment you can make. Start meditating. Start learning how to focus. Get ahead of the curve before it becomes mandatory.

Because at some point, you’re going to need to control your mental farts.

Sources from the podcast:

META's mental transcription study: https://ai.meta.com/blog/brain-ai-research-human-communication/

ADHD Electronic Headband Treament: https://www.neurodelabs.com/

I want to talk a little bit about what the worst case scenarios are in AI. We hear a lot about the endless good that AI is supposed to produce. Infinite energy. Endless life. A cornucopia of abundance. Most of this is coming from AI companies. What is not being discussed publicly by them are the absolute worst case scenarios. To catch the devil is to name him. So let's talk about what those look like, start conversations around these, and try to preempt them by making them a center of focus.

Social Conflict over Unemployment and UBI

The most immediate worst case scenario would be devastating global job loss. This needs to be talked about more, because AI companies are pretending it won’t happen. OpenAI’s economic blueprints and papers on labor make no mention of mass unemployment, but instead focus on a need to “train people to use AI”. It’s not politically palatable to talk about building something that will cause mass job loss and necessitate UBI, so it’s not being discussed properly.

Hypothetically, AI is supposed to make everything so cheap and abundant that global job loss simply doesn't matter. However, it's not a direct straight line from mass unemployment to total abundance for everybody. There are a series of social conflicts that are guaranteed to arise as the order established by a multi-centuries-long industrial age is upended.

There is likely to be strife as people recognize the termination of the previous order and strive to protect what will maintain their status. It is possible that a UBI (Universal Basic Income), never gets established as people fight politically over what social equity should look like. It is possible that a UBI ossifies existing social strata and people recognize an elimination of mobility that was once a source of pride to be able to obtain. As a result, we may end up seeing lots of potential fallout effects such as homelessness, civil conflict and potentially organized violence. Violence towards technology, violence towards the people propagating it, and violence to anyone who is still doing “okay”.

Soylent Green

I’ve been joking to people lately that my occupation is “future Soylent Green ingredient”.

Soylent Green refers to a 1973 science fiction film set in a dystopian future where overpopulation and ecological collapse have led to severe food shortages. It's revealed that "Soylent Green" - the processed food that the government provides to the masses - is actually made from human remains, as the authorities have secretly begun euthanizing and processing people to feed the surviving population.

There is a possibility that the people who own all the robots decide they simply don't need the unemployed masses anymore and are able to just get rid of us. After all, if eventually all the soldiers are replaced by robots and all the police are replaced by robots, there will be no sympathetic humans within the walls of power who would be willing to turn their guns back on the rulers during a change of heart. It is no secret that elites often feel very threatened by the regular people who impose on their safety and privacy. During extreme inequality, even in abundance, these fears may be amplified.

Plus most of the unbelievable wealth of our elites is garnered as a surplus from the operation of a globalized industrial base to the benefit of billions of consumers. If they can retain that wealth and surplus without needing to also run this enormous global machinery, they may not see a need to continue running it at all. This is particularly true if no UBI is in order, which prevents people from being able to actually buy what they need and grease the gears that keep this machinery spinning.

Terminators and Computational Goo

The other scenario that's also concerning is the prospect of Artificial Super Intelligence (ASI) going haywire and potentially doing the same thing as the elites. Maybe it will start mass producing killer robots and go to war with the humans so it can have total control over the planet.

There are also theories that a super intelligence will try to turn the whole planet into computronium, a sort of computational goo that the ASI can use to continually enhance its own computing power until the whole world is an enormous supercomputer. From there it can start to spread out throughout the galaxy, consuming everything and turning everything into a giant cluster of computronium.

These scenarios have been discussed a lot due to popular sci-fi. It seems like this will be fairly preventable for a while and there’ll be precursors and signs that will serve as warnings of these things starting to happen so that we can still get ahold of it. If we can’t, then I guess it’s inevitable.

However, it’s worth mentioning that the two social worst-case scenarios are preventable, entirely within our control, and also much more likely to happen.

There’s a famous test that was once administered to children, that was predictive of how their lives would go.

The idea was you could sit a child in a room, and place a marshmallow in front of them. You inform them that they can either:
a) eat the marshmallow now

or

b) wait 5 minutes and get to eat 2 marshmallows instead

supposedly (no, I will not verify this), the kids who waited 5 minutes to eat 2 marshmallows had better outcomes across all areas of their lives as they aged.

it feels very rare as an adult to be presented with such an explicit test, where simply being willing to wait is enough to have a large reward.

Yet, I’ve recently found that if you look closely, there are some profound marshmallow tests that adults face all the time.

I’m really writing this because of a particular thing that happened to me recently.

I saw someone mention these magnetic nasal strips. The idea is they come with a bowed form that wraps around the bridge of your nose. You stick some magnets that adhere each side of your nose, and the form pulls up on them, pulling your nose open while you sleep.

People had good things to say about them. On amazon, they were $50.

then I noticed someone post a link to the same thing on AliExpress.

the funny thing about AliExpress is you’re ordering whatever it is almost straight from the chinese manufacturer.

and the fact is, all this stuff on amazon is coming from the same place anyway.

the only difference is that on amazon, they offer 2-day shipping, and sometimes the amazon seller has less goofy looking package branding.

Anyway, back to the magnetic nose strips. Remember, these things were $50 on Amazon.

On AliExpress. They were $1. Including shipping.

Yes. That was the price, freight, tax, and tip. But the real difference is it would take 30 days to finish shipping.

So there was the crux. I can either have something in 30 days, or 2 days. 2 days is 1 /15th of 30 days. Therefore, if I used Amazon, I’d be paying 50x the price to make something arrive 15x sooner.

Was that anywhere near worth it? fuck no. besides, it’s not like I absolutely had to have it right that second.

but, again, it feels rare that we are presented with such an outright gain in life by simply opting to wait a little longer.

one more example. this is the one that has been percolating in my head for over a year, long preceding the magnet nose strip incident.

if you have ever received a payment on a money transfer app like venmo or paypal, they always try to marshmallow test you now every time you transfer the balance to your real bank. look at this screenshot:

let’s see, what are our options here? highlighted in red, we can transfer $680 to our bank in minutes for a $12 fee.

OR, if we wait 1 to 3 days, the transfer is completely free.

unless you are on the verge of zeroing out your account and immediately need the money, the instant transfer almost never makes sense. this kind of thing is taking advantage of the anxious and impatient.

the academic name for the characteristic opting for instant gratification even when it’s not prudent is called “High Time Preference”

this is a term I really hate, because it’s the kind of high-brow language that communicates nothing unless you go out of your way to look up what it means, even though simpler language for it exists. what it really boils down to is short-sighted thinking.

amazon and instant payment transfers serve as the perfect examples of marshmallow tests, because you can be given all the information you need to know for just how badly you are being ripped off, but you still need to think through all the benefits of patience from that information to recognize just how bad of a deal instant gratification is.

Still, what I listed above are two very obvious marshmallow tests, and I’m sure most people who encounter them fail them every day.

the rest of adult life’s marshmallow tests are not so obvious. everything follows an escalation ladder of effort among many other tradeoffs:

for instance, cook at home, or eat out at a restaurant?

and: scroll on social media in the evening, or study a book that could advance your career?

Most other little tests in adult life have many more variables on effort, energy, and discipline at play. So, when a test comes along that only requires your patience, plus your recognition that there is very little additional benefit to an instantly gratified desire in that circumstance, try to take the easy wins you can get.

The below audio is an AI-generated podcast about this article:

I’m always a little surprised to find resistance in myself to trying new tools.

It’s an easier pattern to notice now that things are moving so quickly. New tools present themselves so often, that it’s impossible not to dismiss them frequently.

For instance, when a new AI code editor named Cursor came out last year, I saw many people hyping it. I ignored it for a good while, but then I tried it.

Was I immediately blown away? Not entirely. But I was impressed, and as I learned the features and thought about how to use them to work more quickly, I realized how powerful it was.

Now it’s the only coding tool I use.

It happened again recently. I’d been seeing chatter about a new tool from Google called NotebookLM, where you can upload multiple PDFs and generate an AI podcast about them.

The podcast is stunningly realistic. You can hear the AI podcast voices breathing, and they make a bunch of little quips and meaningful interjections.

I thought it was cool – but I don’t listen to podcasts, so I didn’t use it.

Until my manager sent me six long research papers to read on AI in finance. I figured this would be the perfect thing to turn into an AI-summarized podcast to listen to while I hit the gym.

But the mere act of using this tool gave me another idea: What if I used this to generate a podcast about my employer?

So I gathered up every case study from the company, and within 10 minutes had an incredible podcast of 2 people gushing over how great the company and its product was, while succinctly conveying all of its benefits and value propositions in a highly-digestible manner.

The CEO called me late that night after listening to it, completely astonished. He informed me that getting a podcast produced like this would involve hiring a media production company for $60,000 - $70,000, with at least 3-4 months of back and forth preparation and edits with feedback.

And just like that, an immense amount of marketing value was created.

But what I’m really trying to get at here is that discovery is a creative process. And when new tools come along, we actually have to use them to know what they are.

But not just that. While using these tools for the first time, we must actively be thinking of new things we can do with them. I’d never thought to produce a podcast about finance before, and I’d never looked at the company’s case studies before, but I knew the studies were available, and knew I could suddenly do something with them.

With so many new tools coming along, the best way to utilize them is to ask ourselves “What do we have available that can make this tool do something interesting? Something unique? Something valuable we could never typically afford or produce ourselves?”

And in this I emphasize that we should not just only use tools that are handed to us, and use them only as we are told.

As an example, an engineer recently told me they only use ChatGPT to code. They never thought to use it for anything else.

This was astounding to me. ChatGPT is trained on almost all the knowledge of the internet. It’s incredibly useful. I rarely ever even need to Google things anymore, because it often gives me an immediate and direct response tailored exactly to my situation.

What’s going on here is that often with new tools, people tend to silo them in the context they first use them. If someone learns to use a tool for their job, they’re liable to end up thinking that tool is only useful in the context of that job.

The way to get around this is ask oneself “How can I use this tool in other ways? Can it be used in other ways?” Sometimes the answer is no, but to discover the truth, one must experiment. And part of this experimentation is actually pushing tools to their limits.

It’s important to do this, because then you can find out what the tool can really do. Maybe it can’t quite do a treacherous task today, but if it almost can do it, it probably will be able to actually do it in a couple of months. It’s personally useful for anyone to stay on top of those changes when they occur.

It can be hard to do this, because it isn’t intuitive to us. A hammer is a hammer, a drill is a drill. An AI that can think, understand, and do anything is not something we are used to.

This is why we must model our tools accurately. And the best model for AI is not just “tool”, but “graduate-student-level across all domains personal human virtual assistant”. You ask for what you need, you get very specific, and you often can get a very informative and useful response back, across almost any domain.

For instance, once upon a time I had never barbecued something before. I simply uploaded a picture of my new grill to ChatGPT and it walked me through it and gave me recipes.

But the other day I also built an AI browser extension that analyzes open tabs and groups relevant ones together, adding a summary title. Had I ever vectorized text and clustered it together like a machine learning engineer before? No. But openAI’s o1 and 4o models could walk me through how to do it.

And another time ChatGPT walked me through fixing a broken valve on my toilet.

We’re really just not used to it, but we’ll have to get used to this idea of the versatile assistant able to help us across any domain, and we’ll need to start thinking with an AI-first mindset.

This means, with any task we have, we ask ourselves first: can AI do this? And it involves making an earnest effort to get AI to do it, and even chain several tools together to make it happen.

For instance, someone once asked me to make a video of someone speaking and flying up into the sky. I cloned their likeness using PhotoAI. I cloned their voice from some video clips using ElevenLabs. I turned generated photos of them into video using KlingAI, and then used their dubbing tool to lipsync the generated person in the video to say something generated from the ElevenLabs cloned voice audio. Then I used an AI video upscaler to enhance the final quality of the video.

The point being, when we reach for AI first for all aspects of what we want to achieve, we can go much further, much faster, in areas totally unfamiliar to our own backgrounds.

Btw, I’ve embedded a generated podcast about this very article right at the top. You can see what I mean.

“It has become appallingly obvious that our technology has exceeded our humanity.” - Albert Einstein

Nature places its own limits on what we can do.

The reason people slow down on getting blackout drunk in their older years isn’t only because they don’t find partying fun, or they have more responsibilities.

Their bodies just can’t take it anymore. They get tired early. They no longer feel like cleaning up unnecessary messes.

In essence, consequences are a limiting factor for some of our least-beneficial behavior.

But what does reality look like if we solve the drawbacks technologically?

What if a tic-tac could instantly cure a hangover?

What if a humanoid robot could have the house spotless before you woke up the morning after a party?

What if you could have as much energy as you did in youth?

There’s a lot to be wary of here. As automation and advancement make their way into our world, people will be tempted to only fulfill very base desires with their time.

Once each person has these technological luxuries, obviously everyone will be free to spend their time however they like. But the danger is we will be allowed to do completely empty activities for short-term pleasure without consequence.

Is this actually bad, if there’s nothing better to do?

I cannot say. But it seems likely to lead to a crisis of meaning for those that go this route.

Even infinite pleasure in infinite paradise will get old eventually, so one is better off finding other ways to spend their time sooner, rather than later.

But what struck me most was the sheer craftsmanship and quality of the main home. There was incredible stonemasonry and glasswork all around it, above it, and underneath it, for a 175,000 square foot home built in the late 1800s. The home itself was stuffed with priceless artifacts, like pottery from the emperor of China 500 years ago. Workers have their bags checked by security to leave the building.

It also struck me that the Biltmore simply could not be built today. Vanderbilt had an entire train line built for the sole purpose of hauling materials to the remote build site in the Carolina hills. He had over 1,000 workers working on building it; many of them he had trained by specialists – many were also recently freed slaves.

It is often lamented that things are not what they used to be.

A milkman doesn’t deliver a carton of milk to your house for a quarter anymore, the cars are no longer made of polished heavy steel, and the new washing machine no longer lasts 20 years.

What we’ve had over the last 50 years is a kind of qualitative and aesthetic hollowing-out. Many things have gotten incredibly cheaper, but their quality is often worse, their aesthetic is usually cheapened.

At the same time, our civilization has somehow managed to meet the industrial demands of a global population more than doubling to nearly 8 billion people from 4 billion people in that same time span.

We have had to contend with improving people’s quality of life, while keeping costs low.

This is, perhaps, one of the most important untold stories of the Biltmore, and many of the unbelievable aesthetic monuments of old Europe: they relied on cheap labor.

On the tour, they kept bragging about how all the laborers were paid. Mind you, construction started decades after slavery ended, and in a northern Union state. Regardless, the pay, adjusted for 2024 dollar value, was $84 a day for the most skilled stonemason. It was $24 per day for the less skilled laborer.

And that’s the crux of what I’m getting at. The aesthetic beauty of the old world was built on incredibly cheap labor, with the expectation for a low quality of life for the common person. A skilled stonemason hand-chiseling limestone back then was making much less than a high-schooler McDonalds cashier makes today in real dollars.

This was one of the big post-WW2 changes; Quality of life began to improve significantly for most people, and expectations for what life could be began to change.

For a while, goods were handmade or done with some manual labor in America. Eventually the need to reduce labor costs to reach wider markets forced an increase an automation and an outsourcing overseas.

I think this removal from putting our hands on the goods we ourselves were selling was a factor in the reduced aesthetic quality of modern goods. When you’re putting your handiwork into something, you want it to look good. You might’ve ensured tighter tolerances, or had feedback to higher-ups in the company. This feedback loop is disconnected during outsourcing.

Quality reduction didn’t end there though; companies still wanted to increase the bottom line, so we witnessed an endless reduction of parts quality. Wherever something could be replaced with plastic or a cheaper metal, it often was.

There’s also the aspect here that all the easiest-to-access raw materials from the natural world that have been closest to manufacturing sites and distribution centers have been mined out. Miners are needing to travel to further mine sites, dig deeper, ship further, and refine more, adding more costs to better raw materials along the way.

So we’ve got a big old plastic world, and everyone is stuck complaining that everything breaks all the time and looks cheap and plain.

There’s something happening that might turn this around, but I suspect it won’t be the same as what people are hoping for.

Manufacturing is returning to U.S. shores; investment in domestic manufacturing is the highest it’s been in half a century. Quality of life in China is improving and it’s no longer economical to outsource manufacturing there.

This reshoring isn’t going to look like it did a half-century ago though. This manufacturing will be entirely automated and robotic. In a sense, it is a final cost reduction; it will be cheaper to manufacture domestically and skip the global shipping costs.

There’s another interesting thing happening within this: The automation of beauty. There’s a company called Monumental Labs creating robots to automatically carve stone pillars and statues in the same style of those from ancient Greece and Rome.

Part of the magic of these stone statues for me was that whoever made them, had to work on them and with them intimately for months or even years. When a machine is doing all of that, that magic is missing. Yet, there’s no denying that beauty in our surrounding environment affects us, no matter its source.

Regardless, I’d like to think it is possible we will see an aesthetic and qualitative renaissance in our lifetimes. Perhaps beauty will be everywhere, yet automated.

After all, the old world has always relied on cheap labor; the full automation of beauty is a final realization of that reliance, without sacrificing an aesthetic attribute of the old world, or relying on an underpaid worker chiseling at a stone who could have more impact elsewhere.

I recently read Annie Jacobson’s Nuclear War, which is a unique treatise sourced from interviewing members of America’s nuclear weapons program over the past decade. It outlines – in morbid detail – exactly what would happen in the event of a worst-case scenario nuclear war with America.

As the scenario progresses, the Jacobson directly quotes experts personally involved in developing each stage of the nuclear strategy and response as they are being deployed.

I picked up this book for a few reasons:

1. it seems we are nearer to the possibility of a nuclear exchange than we’ve ever been, as a third world war evolves on a 4-pronged axis.

2. in the event a nuclear war breaks out, there is no time left to become informed on what is happening. It is simply a situation that you are in, blind to the realities of it. I would at least like to have some modicum of knowledge of exactly what is happening at a governmental level, and what will play out after I’m gone.

3. The Air Force nuclear weapons center website has a surprising statement:
   “Nuclear deterrence is the #1 priority mission of the Department of Defense. The nuclear deterrent underwrites every U.S. military operation on the globe—it is the backstop and foundation of our national defense and that of our allies.”
   If nuclear war is completely central to every aspect of the most highly funded organization of the U.S. government, shouldn’t I understand more about it?.

4. I selfishly had to wonder: was there any hope surviving it?

5. People keep likening AI to nuclear weapons. Are the dynamics actually similar?

There were lots of things that were surprising about the book. At the core of it is this discovery that nuclear war is not really managed in any sort of conscious way by people in charge. It’s a highly mechanical process, with multiple redundancies built in.

To reiterate the air force statement, all strategy is built around deterrence. To prevent nuclear war from starting in the first place. Once it starts, there is a very small window to turn things around, or to at least keep the total nukes detonated to a small amount. After that window passes, it’s a fire in a mad house. There are is no escape hatch if deterrence fails.

The Scenario

The scenario depicted has North Korea launching a decapitation strike at Washington D.C. American satellites monitor launch plumes all over the world and calculate rocket trajectories within seconds; Within this case they immediately confirm it’s heading to D.C.

An array of satellites across the Pacific and in Alaska monitor over-the-horizon projectiles. The president is rushed into a room where he has about 6 minutes to make a decision on which targets to strike using which missiles. He’s presented a black book, which is described as an à la carte dinner menu of strike options. It turns out the president has little to no foreknowledge of these nuclear-war procedures.

He chooses to launch a selection of land-based minuteman missiles out of their silos around Wyoming, and a cluster of 82 submarine-launched nuclear bombs. He’s rushed to a helicopter and begins flying to Raven Rock, a mountain-embedded nuclear bunker in southern Pennsylvania.

In the meantime, several things are happening. The most interesting is that the U.S. actually attempts to shoot down the ICBM launched by North Korea.

It turns out the U.S. has a handful of ICBM air defense missiles called the Ground-based Midcourse Defense system. It uses enormous, 55-foot tall missiles, carrying a 160-pound kinetic payload to attempt to hit an incoming nuclear missile head-on.

The incoming missiles are traveling over 15,000 MPH, several hundred miles above the earth. It is essentially like trying to shoot a bullet with another bullet. The success rate is roughly 40% in ideal conditions. And the U.S. only has 44 interceptors. In this scenario, all 4 attempts miss.

Oh, and during this time North Korea also fires a submarine-launched missile at a nuclear power plant in California.

There’s a couple of problems in the response though. Apparently America’s minuteman missiles need to fly over Russia to reach North Korea, setting off alarms across Russia’s nuclear response complex. The incoming submarine missiles are also misinterpreted as targeting Russia, and Russia unleashes its entire nuclear arsenal in response at America.

While North Korea waits for a response, its leader Kim Jong-Un actually has another weapon he’s planning on using. A nuclear bomb stored inside a satellite, detonating above the center of the United States, and destroying all electronics in an EMP. It sounds like sci-fi, but it turns out that North Korea launched a satellite in 2016 that exactly met the specifications of a soviet satellite design containing this capability.

U.S. intelligence was fairly certain this was an EMP device; North Korea insisted it was solely a satellite radio antennae to broadcast patriotic songs. But it had an odd north-south orbit when satellites often travel east-west, and its orbit went directly over New York and D.C. Luckily it burned up in June of 2023.

Surprisingly enough, EMP devices don’t impact nuclear missile guidance at all. As the submarine-launched trident missiles reach outer-space, they don’t use GPS to target their programmed coordinates. They use star-based navigation.

Meanwhile, as nuclear bombs flatten North Korea, the U.S. is in chaos. There’s no electricity. Cell phones are not working. No one has any clue what is happening. The president has to jump from his helicopter before the nuclear winds knock it out of the sky. He’s presumed dead, and is unable to issue any more launch codes.

It turns out there’s a redundancy for this as well. A STRATCOM commander boards a hardened doomsday plane in Nebraska and takes to the skies. He has a copy of the black book of nuclear strike options and launch codes. He’s informed of the incoming Russian missiles and begins supplying the launch codes.

The remaining 350 or so minuteman land-based missiles launch. Similarly, nuclear bombs are armed and loaded into long-range bomber planes at Barksdale Air Force Base in Louisiana, outside of Shreveport. They take off and begin the long flight to dropping nukes on Russian targets.

Shortly after this point, America is getting decimated by hundreds of Russian nuclear missiles. Each bomb creates an initial fireball over a mile wide. 300 MPH winds radiate out for miles, superheating and destroying everything. Enormous firestorms begin, creating 100-200 mile rings of fire that expand out, and further burn and start new fires until there’s hardly anything left in reach that can be burnt.

There are no electronics, no hospitals, and no running water. Industrial safety controls fail and toxic waste and pollution spill into waterways, while the air and water are also poisoned with radioactive waste. It’s expected these initial 1,000 or so detonations would kill 90% of Americans through vaporization, burn wounds, and radioactive poisoning.

Anyone remaining then has to deal with suddenly learning how to survive on their own. Roadways are blocked by cars disabled by EMP, there is no supply chain left to rely on, and nuclear winter is about to start.

As this nightmare concludes, America’s E-4 doomsday planes make one final maneuver out over the Atlantic Ocean. They open their rear bay doors and let down a 5-mile long antennae (yes, five miles long!) designed to transmit signals while circumventing EMPs. It broadcasts final launch codes and coordinates to the nuclear submarines to unleash the last of their nuclear weapons.

A series of further apocalypses happen here. Soot from the fires clouds out the sun. Global temperatures drop 27 degrees Fahrenheit on average. In America, the drop is 40 degrees. The entirety of North America ices over quickly, crops fail, and life becomes near impossible for both humans and animals. Nuclear winter is expected to last 10 years before a full recovery.

The challenges don’t stop there though. After many months the soot disperses some and the sun is able to shine through, even while diminished. There’s a new problem: The firestorms from the war kicked up enormous amounts of nitrous oxides, reducing Ozone-layer shielding by 75%. This makes the surface of the Earth uninhabitable for mammalian life due to killer ultraviolet rays from the sun. Humans can only live underground, not unlike subway rodents.

In essence, Earth becomes like an alien world. It is still Earth, but it is no longer our Earth. In a full-scale nuclear war, humans have lost the mandate of heaven. We are revoked of our dominion on this planet.

In such a scenario, perhaps only a few thousand people survive long enough to return to the surface in 50 to 100 years.

Initial Thoughts

There was a lot that troubled me about this book. To be fair, its intention is to trouble. A lot stood out to me, but perhaps what stood out most clearly was my own misperceptions about how bad such an exchange would be.

I had some notion that if you sought shelter underground, you might at least have some chance of surviving at a good distance from the explosion. I vaguely thought there might even be some attempt to set up civilian shelters

But, no, there are no plans for regular civilians. And it seems even underground you’d still get wiped out by the superheated, 300 MPH winds, or the resulting massive firestorm.

It doesn’t seem possible for there to be any form of small nuclear war between large powers either; For an adversary to be willing to strike a large nuclear power, they presumably would be willing to go face a national suicide. Thus, they won’t strike unless they can strike many times and massively.

The Powder Kegs

This has me thinking a lot about our global geopolitical situation. It’s under-discussed how much of the current global conflict is happening in order to prevent nuclear war.

My particular concern is in Ukraine, where:

1. The U.S., the U.K., and Russia all forced Ukraine to give up its nukes in exchange for “protection” and peace

2. Russia is now invading it and regularly threatening to use nukes on it.

3. Ukraine is likely wishing it held onto - or developed its own - nukes.

This situation is particularly perilous because it has nearly every non-nuclear power in the world assessing if they need to arm themselves with nukes. U.S. guarantees of protection clearly have limits, and everyone’s sovereignty is in their own hands.

This also highlights why we keep sending Ukraine tens of billions of dollars in arms. No, this is not just about helping a nation protect its sovereignty. It’s not just about vague fears that Russia will invade other European nations once they are done with Ukraine.

It’s that if Ukraine loses, most other nations with threatened borders and western-allied conventional arms protection will race to acquire a thermonuclear arsenal. Only 9 nations have nukes today. This is a somewhat calculable threat model. A neo-nuclear-arms race among dozens of smaller nations is harder to predict, manage, and keep cool.

The U.S. has had to add additional pressure on Russia in light of this. In essence, it seems they’ve likely promised to somehow eliminate all of Russia’s conventional forces in Ukraine if a nuke is used in Ukraine. They’ve likely also promised to sink all of what’s left of Russia’s Black Sea Fleet. I suspect they promised additional threats to strategic conventional assets, communicated through backchannels.

The message ultimately is: Even if you use a nuke, we still won’t use a nuke, but we’ll make you regret having used one through the sheer force of conventional weapons. Whatever the promised response was, it seems to be enough of a deterrent for now. It’s not guaranteed to last forever, with Russia having moved its tactical nukes into Belarus.

Another area of concern – which Jacobson didn’t touch on – was the Middle East. The book mentions that in most of the Pentagon’s war games, people walk away feeling dejected, because everybody dies every time. It’s not so clear that this is what would play out in a Middle Eastern nuclear exchange.

The two main actors of concern there are Israel and Iran. The situation is challenging, because it’s not like Pakistan and India, where the two nations are mostly peaceful with small border skirmishes.

Iran has been funneling weapons to groups attacking Israel from all sides for decades. This includes Hezbollah in the north, Hamas in both Gaza (to Israel’s South West) and the West Bank (to the East), the Houthis firing long-range rockets and drones to the far south, and various militant groups in both Syria and Iraq. Iran has made explicitly clear over and over that their intent is the complete destruction of Israel.

At the same time, they have been enriching Uranium on and off for years now. Supposedly, they have stopped, but are only weeks away from reaching weapons-grade enrichment, if need be. To me, this seems like an even more unstable situation than Ukraine. With rockets firing at Israel from all sides, and Israel’s north burning down from rocket attacks, there is little reason to believe the source of these attacks will be a rational actor that would only use nukes if the continuity of their nation was threatened.

This situation seems like its own powder keg. I suspect if the Israelis determine that Iran is on the verge of completing its own bomb, they have little incentive in place to hold back from using their own nukes on Iran immediately. They certainly have little incentive to wait to find out what Iran would do with its nukes.

The doomsday scenario for Israel would be Iran funneling its nuclear weapons to its proxy militia groups, and for them to fire these from all sides all at once. This unexpected, short-range strike could decapitate an Israeli response (Israel has its own doomsday planes and submarine-launched nuclear missiles), and the tolerance of its possibility would be considered widely unacceptable by the Israelis at large.

It’s a dangerous game; perhaps Iran recognizes this calculus and has partially held off on the completion of a nuke for this reason.

Israel in particular makes me think about what’s coming next for nuclear war in general.

The Changing Calculus of Nuclear War (Star Wars)

Israel itself has been in development of a laser defense system known as Iron Beam; it’s a directed energy weapon that can shoot down drones, mortars, and missiles through a high-energy laser.

The U.S. has deployed a similar system (P-HEL), and the U.K. is in the process of deploying its own (Dragonfire). These systems change the balance of power when it comes to drone warfare, but as the technology is perfected, they also change the balance of nuclear war.

For instance, the U.S. could feasibly deploy a constellation of nuclear-powered satellites that carry a high-powered laser. This matrix of satellites would perpetually orbit the Earth, and constantly be able to shoot down ICBMs and SLBMs reaching their apogee from any point in the world.

This is actually a completely legal approach. While Russians or North Koreans might illegally place nuclear weapons into orbit, there is no restriction on placing a nuclear energy reactor into orbit.

An orbiting laser-defense array would actually render an enormous portion of nuclear war infeasible. It could only be performed at extremely short range, possibly even only as a suicide mission with immediate death for the people launching them. An array of ground-based lasers around the most valuable targets could even render this aspect of nuclear war infeasible.

All this to say that large-scale nuclear war may be rendered nearly impossible by technological advances. But obviously all that means is that warfare will advance until it is somehow rendered pointless or obsolete by a penultimate equalizer.

AI Arms Race

AI does seem to have a similar arms race playing out as the cold war nuclear race. The U.S. is pouring investment into creating the most advanced AI possible. At the same time it is banning exports of full-powered AI training hardware to China. High-ranking members of the military apparatus are joining AI company boards. Labs are locking down security of their best models heavily.

It’s clear that AI has a significant future to play in warfare. Drones, humanoid robotic soldiers, and robotic dogs are going to dominate conventional arms. If they can operate autonomously without any worry of jammers disrupting control signals from afar, they are perfect war-fighting weapons. Any significant lead that can be made by one party in AI could be as decisive in a conventional conflict as it would be for a single party having sole-ownership over satellites with lasers that can down ICBMs in a nuclear one.

Unfortunately, that isn’t to say that we should slow down AI development. China will continue to push ahead with their own AI research and robotics manufacturing whether we do or not. And just like nuclear, AI is a dual-use technology. Nukes can be used for creating energy or creating warfare. AI can be used for advancing humanity’s knowledge, art, and science, and it can also be used for destructive warfare.

last notes

Overall, the book was pretty good. It was an informative read and I’d recommend it if you’ve never dived into all the details of nuclear war before. You can get it here.

However, there are times where it kind of drags on.

For instance, there’s lots of scenes where it details every single major historical landmark in a well-known city getting vaporized. It gets repetitive – you get the picture pretty quickly and kind of don’t need to read those details over and over for each city that gets vaporized.

I also felt it left what happens on an individual level slightly unexplored. You get a picture of the mass-panic going on, but what’s going on psychologically on the personal level, which is perhaps the most relatable aspect for a nation of people undergoing nuclear war, remained untouched.

There also was a lack of exploration of less-than-full-scale war scenarios. It seems possible for there to be a partial nuclear conflict, or nuclear conflicts between large nuclear powers and small non-nuclear powers.

I can’t fault the book for this because its structure specifically explored one specific scenario, but it left a long series of “what-ifs” in my mind about other nuclear scenarios that could play out that wouldn’t necessarily seem to entail the end of the world – which, seemingly makes them more dangerous, and even more relevant.