I’m proud to introduce Death Clock, a life expectancy calculator that predicts the day of your death based on 17 personalized variables.

Give it a try for free!

This is not only the first “app” I’ve ever created myself, but also my first experience with AI-assisted coding, which has become known as “vibecoding.” 

I can definitively say this was one of the most eye-opening, impressive encounters with technology I’ve ever had, and it opened my eyes to a vast horizon of possibilities that I think AI-assisted coding will open up going forward.

Here are my observations and insights based on my first few hours of vibecoding.

Finding an entry point

One of the first things I look for when trying to learn a new skill is a good “entry point.”

Simply typing the topic into Google or YouTube is not smart, as it only results in a flood of mediocre-quality, clickbait-driven results. Instead, I look for a single, in-depth piece of content or a course taught by a qualified instructor with a strong track record of quality. Ideally, there’s also an accompanying community or discussion forum associated with it, so I can see what others are doing.

All these criteria were fulfilled when I signed up for my friend Nat Eliason’s new course, Build Your Own Apps with AI. One recent Saturday morning I started watching the instructional videos while my wife and the kids were sleeping in.

The course recommends the coding program Cursor. As I downloaded the desktop app and created an account, the complex-looking interface was already starting to intimidate me. I had downloaded IDEs (Integrated Development Environments) before, but since I have virtually no technical ability or knowledge, I always felt overwhelmed and confused by the unfamiliar interface and quickly gave up.

Here’s what the interface looks like in Cursor:

As I watched Nat’s videos, it soon dawned on me how incredibly straightforward and easy it was to create functional code using Cursor, for one simple reason: it draws on the formidable powers of leading LLMs.

As I began experimenting, I found that it wasn’t necessary to interact with any part of the software except for one: the ongoing chat with the AI. Although you can directly manipulate the code, navigate around the various project files, approve or reject individual edits, and issue commands in a terminal window if you want, none of that is strictly necessary. The AI can take all the actions itself.

Building Death Clock v1

After watching the first module of the course, I decided to go right ahead and try building something I actually wanted: a life expectancy calculator. 

I’m in the midst of writing my new book, on the art and practice of completing an annual life review, and I’m finding that one of the main themes of the book is reckoning with one’s mortality and limited time on the planet. It’s quite hard to take on that longer-term perspective day to day. But at least once a year, I think it’s incredibly valuable to do so. I was looking for a way to give readers a visceral, felt sense of how short life truly is.

I can write thousands of words about mortality and its ability to put our lives into proper perspective, but no volume of words compares to the power of a personalized, interactive tool that calculates your own expected date of death. It’s the difference between generic advice and an expiration date that takes into account your own unique life circumstances.

Living in Mexico has also exposed me to a very different attitude toward death than I’m used to in the US. Instead of something to be feared and the mention of it avoided, death is a much bigger part of the culture, from the ofrendas of Día de Muertos to the ubiquity of skeleton iconography to the constant awareness of ancestors. I want to make death something that people can talk about more openly.

And lastly, I turn 40 next week, so let’s just say that mortality is on my mind

Working with the AI

I started my project by creating an empty folder, loading it up into Cursor, and asking it to “Build a web app that calculates someone’s life expectancy based on lifestyle factors.” That was it. There were no technical specifications, no feature requests, and no tech stack. Just a simple request made in natural language.

Cursor gives you the option to select which LLM you want to use, including all the leading models from OpenAI, Google, and Anthropic. I kept it on “auto-select,” which means it decided which model to use depending on the situation.

It proceeded to immediately build a full-fledged React project, which took several minutes and involved creating an entire structure of multiple interlinked files. I asked it to open the app in my web browser, and it prepared the file, which I just needed to double-click.

Here’s what that very first version looked like:

So the app now existed, but with one tiny problem: there was no way to input any personal information! I asked it to add input fields for each of the lifestyle factors, and in a couple of minutes they appeared. I put in my own information as a test, and it spit out the results:

After only about 5 minutes, I already had a functional web app! This alone is astonishing, as I would have probably needed hours to get even this much up and running on my own. I thought back to my early web design efforts using Microsoft FrontPage as a teenager in the 90s, and how that experience was so daunting that I gave up on the endeavor altogether.

I spent the next several hours adding features, which amounted to no more or less than asking for them in the AI chat window and waiting. Here’s what I changed, one element at a time:

• Added a total of 17 questions, encompassing various aspects of a person’s life that influence their life expectancy
• Added an in-line BMI calculator, based on a person’s height and weight
• Asked the AI to create a logo of an hourglass and place it at the top of the page
• Had it try out a couple of styles, settling on a muted, grey and blue color scheme
• Added some interesting outputs and observations related to life expectancy to the “results” page, based on the answers to the questions
• Added a pie chart visualizing a person’s remaining lifespan (which required Cursor to download and install a new library, which it handled on its own)
• Added a call to action to check out my website at the end
• Set up a repository on GitHub and published the app using GitHub Pages, so I can share it with others

All this took about 3-4 hours, but I would estimate around half that time was spent waiting for the AI to do the work, or for the new version to deploy, so I could view it in a browser. While all this was happening, I hung out with the kids, washed dishes, took a shower, and got some reading done. Instead of requiring an all-encompassing, immersive day of obsessive learning, which is what I would have expected, I actually had quite a relaxing, balanced day!

There were only a couple of hiccups I encountered during this project. At one point, there was a bug that would make the whole screen go blank when I clicked the “Calculate Life Expectancy” button. Cursor asked me to copy and paste the error message from the console into the chat, but before I even did that, it guessed what the problem was and proceeded to fix it by itself. Here’s what that interaction looked like:

I sometimes had to perform actions beyond just clicking “accept all,” such as typing “y” into the command line to allow it to proceed, or typing in my computer’s root password to allow it to install something, so I did have to pay a bit of attention and look out for next steps.

A few times, a new feature didn’t work correctly the immediately, and required follow-up fixes, such as this first attempt at a pie chart:

One of the most surprising parts of working with an LLM in a coding environment is that the AI has context and knowledge about many, many domains that have nothing to do with code. For example, at one point, I got stuck with the GitHub settings, so I just pasted a screenshot of what I was seeing into Cursor, and got a precise diagnosis of what was wrong, plus step-by-step instructions on how to fix it.

But it also goes beyond software. At one point, I asked Cursor to tell me how robust or evidence-backed a question about income level was, and in seconds, it pulled in an academic paper that examined exactly this question. It still had to simplify the correlation between income and life expectancy into a multiple-choice question, but knowing there was at least some validity to this question gave me the confidence to move forward.

I kept assuming I would hit a roadblock and have to switch to a “real” LLM to do more serious research, planning, or structuring of questions. But that wasn’t the case: the LLMs that Cursor is drawing on are exactly the same ones you’d have access to via ChatGPT or Claude, so there’s no need to leave.

That also means that Cursor can do design work. I asked it to create a visual style reminiscent of the blocky, 8-bit graphics of early Nintendo games, but with a death-themed, macabre look. It instantly understood the assignment, using red and black to convey the right feeling:

I decided that look wasn’t quite right, so I had it pivot to a cleaner, more modern style. Here’s what that interaction looked like, replacing probably hours of work:

The ability to pivot on a dime like this and effortlessly try out a completely new direction is astonishing. Small experiments and whimsical curiosities can be indulged and tested without expending significant time or effort.

My 4 takeaways from vibecoding

This first experience of a few hours of vibecoding left me with 4 takeaways:

• Learning to code is now optional
• The new bottleneck is how you spend the time that’s saved
• Software is the new frontier of book publishing
• We’re entering a more impressionistic era of creation

Learning to code is now optional

Over the years, I’ve considered whether I should build an app many times. I live in the world of tech and software, and in many ways, the ability to create a new software tool is the pinnacle of agency in that world.

Yet every time, I’ve decided not to pursue building an app, either because I lacked the time to learn it myself, or didn’t want to spend the thousands of dollars it seemed to require to even build something basic.

But now, in a matter of hours and at almost zero cost, I can build something that’s genuinely useful. I didn’t learn anything about coding, but I think that’s overall a good thing. Learning can be fun and is certainly useful, but it isn’t always inherently good or necessary.

Why should the ability to leverage software be limited to those willing to spend months or years studying arcane details of technical implementation? Why should someone’s vision or mission require them to know the low-level details of how a webpage gets rendered? And now, nearly all of the details are “low-level.”

Learning coding is now optional, but I think AI tools will also make it easier to learn to code for those who decide they want to. At any point, you can bring in context from any part of the codebase and ask the AI to explain it to you at any level of detail you want. You can even have it explain things outside that environment, such as the many external systems and interfaces you’ll need to get a full-fledged website working.

This is such a remarkable level of accessibility for a technology that was previously very hard to use, and it’s difficult to predict how the world will change when everyone can wield the power of software.

The new bottleneck is how you spend the time that’s saved

It’s so fast to create and edit code this way that the bottleneck starts to become how fast you can move your mouse, the speed of your internet connection, how long it takes to deploy a new build, the speed of refreshing the page, etc.

I predict we’ll see a variety of efforts to speed up every little step involved in coding, the same way that factories once invested millions in reducing the time it took to switch a production line from one activity to another, as that became the bottleneck.

But more broadly, the true limiter on the quality of software that people will be able to produce using Cursor and similar tools is how they spend the time that AI frees up for them. 

You could spend it chilling by the pool or watching TV, but you have to remember that everyone else is also having all their time freed up, so the level of competition will increase like a rising tide. Many web apps that people will build this way are hobbies, or experiments, or complements to other projects. Still, many will have some kind of competition or alternative, and the only way to compete effectively will be to invest the time saved in new dimensions of quality.

Maybe you spend that time exercising and meditating, so that you can ground yourself and bring wiser, more holistic decision-making to the AI. Maybe you spend it reading and researching, so the knowledge underlying your app is richer and more nuanced. Maybe you spend it hunting for obscure sources or offline archives, so that you can incorporate context that the LLM doesn’t already know. Maybe you spend your time talking with potential customers, so your choices more accurately reflect what they want.

All of these are valid choices, and they will all become important dimensions of competition and quality, even more so than they already are today. The true scarce resource continues to be the time and attention of other people, and I only expect the battle for that attention to keep heating up.

Software is the new frontier of book publishing

One of the domains I’m most excited about applying these new tools to is book publishing. Books have changed so little over time, and increasingly suffer in comparison to other, far more interactive and engaging forms of media.

I don’t think bemoaning this fact and lecturing people on the importance of reading is helpful, but I do believe interactive web apps like this could make a tremendous difference. What if, every time you finished a book, or even a single chapter, you were presented with a link to a free, interactive, personalized web app that directly applied the ideas you just read to your own situation?

Instead of trying to guess how to apply a book’s ideas, or get upsold to a course, or have to get expensive support from a coach or consultant, you would have a self-serve piece of software you can immediately engage with.

The value of a book is that the author has taken an extraordinary amount of time to research and think deeply about an important issue, topic, or skill. That’s a rare thing in our hype-driven online world of disposable headlines. But that same slow-moving, timeless quality makes it very difficult for books to recommend or prescribe any given form of implementation. There’s just too much variation between individuals to offer a one-size-fits-all solution, and long publishing timelines mean that any solution printed in the pages of a book is likely to be obsolete by the time it hits the shelves.

This is a way to combine the best of both worlds: to deliver the timeless, wise, holistic wisdom of books, accompanied by a suite of personalized, customizable, up-to-date digital implementation tools, accessible in one click or tap. This is how you save the culture of reading – not by resisting change but by embracing it. I plan on making extensive use of this approach in my next book.

We’re entering a more impressionistic era of creation

One of the most continuously surprising aspects of AI-assisted vibecoding is how brief, imprecise, informal, and vague my instructions can be, and still be understood. I could almost always just say “Fix this” with a screenshot, or “Make this look better,” or even just “Improve the question,” and AI would figure it out.

This is so different from past technologies that require you to be extremely exact, specifying what you want with mathematical precision. Even a single wrong character in a codebase of thousands of lines could result in a catastrophic error.

My unclear instructions often resulted in better results, because the AI would misinterpret my intentions and make improvements I hadn’t even thought of. Some of the best ideas came from the AI, either because I asked it for ideas or because it contributed them spontaneously as it guessed what I was trying to achieve.

Like the transition in painting style from the Realism of the mid-19th century to the Impressionism of the late 19th century, driven by painters’ desire to capture changing qualities of natural light, fleeting moments, and spontaneous experiences using quick, expressive brushwork rather than carefully finished compositions, we’re going to see a similar transition in software design.

Interestingly, what sparked the transition back then was technology – the advent of photography meant that scenes could be captured with nearly perfect realism, which devalued that ability by humans. Human artists pivoted in reaction, exploring a new frontier of perception, novel color combinations, and everyday life.

I think we’ll soon see our technological creations becoming much more impressionistic, based on ambiguous premonitions, subtle feelings, or vague notions that we can’t fully articulate. We’ll see people create various kinds of software as artistic expressions, or to capture a fleeting memory, or to convey a single message. Software will become its own mode of creative expression for a much wider range of people now that the price of entry has plummeted to near zero.

4 tips for using Cursor

Despite my lack of commitment to learning anything in particular, I found that I did end up learning a few things about how to work effectively in Cursor:

• You don’t have to deploy a new version with every new feature you build, as that takes a few minutes. But it’s a good idea to do so anytime you get a major new feature working, as you’ll be able to “roll back” to that point if you mess up anything in the future.
• It’s always helpful to bring in the relevant context to any interaction with the AI chat. Cursor makes this very easy by including an “add to chat” button both in the code window and in the terminal, which are the two places you’d want to draw on for context
• Pasting screenshots into the AI chat is remarkably helpful, as it allows the AI to see exactly what you’re referring to. Often, you don’t even need to say anything – the AI understands what’s working as soon as it sees how it’s appearing.

You can check out Cursor with a free trial at https://www.cursor.com. I signed up for the paid version for $20 per month to build my first app, though you’ll have access to a lot of functionality for free. I recommend their “getting started” documentation to learn about the basic features, which are more than enough to allow you to build your first simple app.

Follow us for the latest updates and insights around productivity and Building a Second Brain on X, Facebook, Instagram, LinkedIn, and YouTube. And if you’re ready to start building your Second Brain, get the book and learn the proven method to organize your digital life and unlock your creative potential.

I recently came across a Substack post by Venkatesh Rao called Oozy Intelligence in Slow Time that was one of the most insightful I’ve ever read for understanding the nature of Artificial Intelligence.

We tend to think of Artificial Intelligence as being in an arms race with Human Intelligence. Which one is smarter? When will AI surpass us?

But there is a hidden assumption buried in that comparison: that intelligence is best measured by its peak moments – the flash of brilliance, the sudden epiphany, the intellectual breakthrough.

Rao suggests that we look instead at a completely different aspect of intelligence: how much it costs.

Human intelligence is tremendously costly. Most of our time is spent simply maintaining this high-performance machine we call our body. Eating, drinking, sleeping, grooming, socializing, resting, etc. can all be seen as “overhead costs” needed to merely keep us alive.

Because it costs so much just to function each day, every minute of focused attention we spend requires a certain return-on-investment to be justified. We are constantly making choices to maximize that “return” on our attention: Do I spend the next 30 minutes working out, or cleaning the kitchen? Do I spend today working on this project, or that project?

Activities that don’t meet our threshold for “required return” don’t get our attention, plain and simple. This can be understood as a kind of “minimum wage” that our brain must earn, otherwise it refuses to work. 

We make this calculation fairly seamlessly any time we consider engaging in an activity:

• We might be willing to spend 10 minutes reading an interesting article, but not if it takes 30 minutes (“too long, didn’t read”).
• We might be willing to drive 20 minutes to eat at an exciting new restaurant, but not if it takes an hour.
• If you buy an appliance for your home that costs $20, but you spent 2 hours reading reviews and evaluating the options, the return on that purchase is lower than if you had instead received and acted on a trusted recommendation from a friend

This is part of what makes learning anything new so challenging: you have to spend lots and lots of time, with little return on that investment, in order to gain some future reward that isn’t even guaranteed. It’s akin to investing a lot of money into something without knowing if it will ever pay you back.

Another way of defining the “minimum wage threshold” for our brains is patience.

Rao asks: “How often are you in the mood to do boring, tedious, bureaucratic tasks (such as filling out forms, doing your taxes, or opening postal mail)?”

His answer, and mine, is: not often. It’s not that I don’t have the time for such tasks. It’s not that I’m not smart enough or don’t know how to complete them. The problem is that they require too much patience (i.e. they fail to meet my brain’s minimum wage threshold). I thus “can’t afford” to spend my attention on them, and instead tend to put them off for as long as humanly possible, usually until some catastrophic consequence becomes threatening enough that I have no choice.

If you think about it, there are many such tasks that would produce immense benefits for us, if we just had the patience to do them:

• Spending hundreds of hours learning a new language or how to code
• Reviewing every note you’ve taken over the last few years for buried ideas or insights
• Organizing all your personal contacts in a searchable Notion database

These kinds of tasks involving collecting, organizing, summarizing, formatting, and reviewing information would be tremendously valuable if we did them, but often fall into the “requires too much patience” category for most people.

This is where AI becomes so powerful. AI effectively lowers the patience threshold to almost nothing. There is no task that is too boring, too mundane, too repetitive, or “beneath its dignity.” Unlike us, grinding away on such tasks doesn’t annoy it, ruin its motivation, give it a bad attitude, or make it angry at us. It is a dutiful employee requiring a minimum wage of virtually zero.

This is a very different way of understanding AI’s value. It’s not AI’s superintelligence or blazing speed that make it valuable to us: it’s AI’s patience in completing an endless series of tedious tasks that are too far below our patience threshold for us to justify doing at all.

Our attention is expensive, and thus can only be spent on activities with a clear outcome that can be achieved in a predictable amount of time. Whereas AI has an almost infinite amount of attention that is so cheap it can be spent lavishly, even wastefully, on activities that would never be worth our time. We can afford to spend this newly abundant form of intelligence on tasks that are below the minimum wage our brains are willing to work for.

The two components of patience

If we stopped here, this would mean that the main use for AI is completing our boring to-do lists for us. But there’s a level deeper to consider, because patience has two components that can be separated: time and detail.

As you toil away filing your taxes, for example, there are two factors that determine how much patience it takes: the level of detail that you’re required to process and the time it takes to do so. It is the combination of many complex details you have to process over a long span of time that makes taxes so excruciating.

The crucial thing to understand is that we have a minimum AND maximum threshold for BOTH the time we’re willing to spend and the number of details we’re willing to process:

• If it takes too much time, we get impatient and opt out (think of a movie where not enough is happening to hold your attention)
• If it doesn’t take enough time, we get overwhelmed and opt out (think of a short-form video that is so fast it’s aggravating to watch)
• If it presents too much detail, we get frustrated and opt out (think of a book going way too deep into a technical topic you don’t understand)
• If it presents not enough detail, we get bored and opt out (think of a children’s book with not enough complexity to be interesting to us)

In other words, as humans, we have a clearly defined “window of attention” that limits what we’re able to pay attention to for long periods. Our attention span is an actual span with clear limits. Anything outside of that – that either moves too slowly or too quickly, that demands too much of our brains or too little – is tremendously expensive for us to attend to.

When you say “I don’t have patience for that,” you’re not saying you don’t have enough time. You’re really saying “That is below the level of detail I can sustainably process at the required rate.”

Thomas Carlyle once said, “Genius is an infinite capacity for taking pains.” The word “pain” is informative in this sense – it is actually painful for us to stay outside our window of attention for long. The brain has a “focal length” it is comfortable with, like our eyes, and that attention span evolved for chasing and hunting down an animal about our own size, or picking fruit from trees; anything outside of that feels unnatural and painful for us.

This might seem like a discouraging and even fatalistic view of human potential: there are just some things we can pay attention to, and some we can’t.

But there’s another detail that Rao highlights here: that time and detail interact and influence each other. They are not independent variables: changing one actually changes the other.

Consider that looking at a raindrop with your naked eye might be boring, but if you zoom in with a microscope, you’ll see millions of microorganisms blooming and buzzing in stunning diversity. The classic example of “watching paint dry” would be terribly exciting if you could zoom in to the molecular level and watch the symphony of chemical reactions playing out.

In other words, the more you zoom in, the faster things are happening. The rate at which time passes for you depends partially on how much information you can take in. It’s not that time is actually speeding up or slowing down – your perception of time is speeding up or slowing down, and that perception is strongly influenced by how much detail and complexity you can take in and process.

Considering this idea, perhaps it is not people’s intelligence that limits what they can pay attention to and learn: it is their patience. And what limits their patience is not some stoic quality of their character, but their ability to zoom in and take in enough detail that reality feels interesting.

This also changes our view of what exceptionally patient people are doing. It’s not that they have some inner reserve of steely endurance – it’s that they’re better at operating at a level of detail where things happen faster.

The gardener absorbed in the intricacies of trimming a bonsai tree, or the basketball player shooting hundreds of free throws in one practice session, or the chess grandmaster playing through dozens of alternatives of a match – maybe these people aren’t abnormally patient; they’re just better at zooming in to a level of detail in their craft that the full bandwidth of their attention can be occupied.

We tend to think of patience as primarily a moral virtue, alongside work ethic, honesty, integrity, and empathy. What if instead we removed the moral framing, and thought of it instead as a side effect of the way we consume information?

AI could be used to tweak and tune information with the goal of fitting it into our preferred window of attention. Instead of treating the content we consume as one-size-fits-all, we could use AI to modify that content so that it’s at the right speed and the right level of detail such that it feels captivating and enlivening for us to pay attention to.

If a piece of content is too detailed, we can ask AI to summarize and distill it for us in ways that a novice can understand. If a piece of content is not detailed enough, we can ask AI to elaborate and add more sources and examples.

If a piece of content is coming at you too fast, you can ask AI to slow it down, break it into chunks, and give it to you one piece at a time. If it’s coming too slowly, you can ask it to move faster and progress to more advanced topics sooner.

I can foresee a future in which we rarely consume a given piece of content without changing it to suit our preferred window of attention. A future in which we run all our content through an AI curator who refines and modifies it to fit how our brains work. Not doing so will feel like buying a pair of shoes without trying them on for size.

In that future, patience won’t be considered a moral virtue – it will be considered a failure to properly utilize the tools at our disposal to customize our experience according to our needs.

If you’d like to read the Substack post by Venkatesh Rao called Oozy Intelligence in Slow Time yourself, click here. I can also recommend Matt Webb’s article Intelligence Too Cheap to Meter on this topic.

Follow us for the latest updates and insights around productivity and Building a Second Brain on X, Facebook, Instagram, LinkedIn, and YouTube. And if you’re ready to start building your Second Brain, get the book and learn the proven method to organize your digital life and unlock your creative potential.

Like so many others, I’ve spent the past year exploring and experimenting with emerging AI tools. 

Throughout that time, there has been one question I’ve been trying to answer: Will AI replace the need for Second Brains entirely?

A lot of people seem to think so, and I admittedly have a self-interested motivation: to decide whether I should continue advising people to build a Second Brain at all, or just tell them to rely on AI and save all that effort. 

After many dozens of hours of experimentation, my conclusion is that AI is not going to replace the need for a Second Brain anytime soon.

Here’s why: no matter how powerful AI becomes, the data we put into it has to come from somewhere, and the AI’s outputs have to go somewhere. A Second Brain (or whatever you want to call it) is still needed both as the repository of all those inputs and as a staging area for storing those outputs until they’re ready to be used.

What’s Changed – Organize and Distill

There is no doubt that AI is going to radically change what we think of today as the creative process.

Looking at my CODE framework representing the creative process, however, it is mostly the middle stages of Organizing and Distilling that AI is transforming.

Organizing (step #2) is the stage of the creative process that inherently adds the least value – it is only needed to prepare the ground for the subsequent stages. Thus it’s no surprise that it’s the first one to be automated by AI. 

No longer does it make sense to meticulously format your data in a perfectly organized database – instead you can just dump a morass of text into a prompt window, and AI is smart enough to understand what you intended. 

As an example, Notion has added AI to its software, allowing you to interact with and “talk to” your notes without having to spend a lot of time adding structure.

Distillation (step #3) is also a perfect fit for the rapid, emotionless decision-making of AI. Large Language Models excel at rapidly summarizing huge amounts of text at whatever level of detail you desire.

For example, in my video on using ChatGPT to summarize books, I showed how AI was able to save me dozens of hours of formerly manual work to end up with a concise, actionable book summary.

What Hasn’t Changed – Capture and Express

The first stage of the creative process – capturing information in the first place – has still hardly been touched on the other hand.

New apps like Rewind allow you to record everything that happens on your computer, but in my experience that just creates a lot of recordings to wade through.

Although some capture tasks like digitizing handwritten text have been automated, we still have to write down our thoughts and ideas in the first place!

The quality of an AI chatbot’s response is always dependent on the quality of the inputs you provide it. AI cannot (yet) go out into the world and collect its own data, so we have to do that ourselves by capturing notes, highlighting passages in books, taking pictures, and saving our favorite ideas.

The fourth and final stage of creativity, expression, also still requires a human to decide what to do with the outputs of ChatGPT and other AI tools. Someone has to put the finishing touches on the final product via their own voice, style, taste, or perspective.

My wife Lauren’s video about creating a children’s storybook using AI perfectly illustrates this point: although every major component of the final product was created by ChatGPT, it was Lauren’s direction, synthesis, and creative nudges that allowed all the parts to come together in a cohesive, meaningful whole.

AI Concentrates Human Creativity at the Initial and Final Stages

AI doesn’t make human creativity unnecessary – it concentrates our creativity at the beginning and end of the creative process.

For a concrete example, in my video on Google’s new AI platform NotebookLM, I demonstrate how I can import the entire history of my reading highlights, and then freely make associations and connections out of that vast collection of text totaling 594,379 words from 719 sources.

While that capability seems almost superhuman, notice what it still required of me: to do the reading in the first place and save the excerpts I found valuable (capturing), and then to take NotebookLM’s responses and turn them into my own creation (expressing). In other words, the first and last steps of creativity haven’t been touched.

I can effectively skip from the first step to the last step, barely touching the steps in between. But that means I still need to take the first and last steps, to give the AI a starting point and an endpoint.

The relevant question has become: what do we do now that the “cost” of intermediate steps like organizing and distilling has plummeted?

Tasks that formerly required expensive human effort can now be completed with cheap computer effort, in fractions of the time. What kinds of goals, outcomes, and creative projects have suddenly become far more feasible than they were just a couple years ago?

For an example of what it might look like to work with AI as a real-time creative partner in this way, check out my in-depth interview (Part 1 and Part 2) with Srini Rao on the AI-powered noteaking app Mem (which by the way is the only notetaking app that OpenAI has invested in).

Follow us for the latest updates and insights around productivity and Building a Second Brain on Twitter, Facebook, Instagram, LinkedIn, and YouTube. And if you’re ready to start building your Second Brain, get the book and learn the proven method to organize your digital life and unlock your creative potential.