2025-09-11 00:23:24
Hi all,
One of the enduring pleasures of writing Exponential View is returning to the thinkers who first taught me how to see the world. Their essays, written between the 1940s and the 2000s, still frame how I make sense of technology, systems and society.
Today, I’m going to share ten of these essays with you. I’ve grouped them into three clusters:
Essays that sharpen our intuition around exponential trends,
Essays that help us think in systems,
Essays that train us to recognize patterns across history.
With each one, I’ve added a brief note on why it matters to me – and why I think it matters to our community.
Enjoy!
The greatest shortcoming of the human race is our inability to understand the exponential function.
Why it’s relevant: I include Bartlett because his deceptively simple framing on exponential growth still cuts through today to remind us why so many forecasts and intuitions fall short.1
[B]ecause we’re doubling the rate of progress every decade, we’ll see a century of progress–at today’s rate–in only 25 calendar years.
When people think of a future period, they intuitively assume that the current rate of progress will continue for future periods. However, careful consideration of the pace of technology shows that the rate of progress is not constant, but it is human nature to adapt to the changing pace, so the intuitive view is that the pace will continue at the current rate. […] From the mathematician’s perspective, a primary reason for this is that an exponential curve approximates a straight line when viewed for a brief duration. So even though the rate of progress in the very recent past (e.g., this past year) is far greater than it was ten years ago (let alone a hundred or a thousand years ago), our memories are nonetheless dominated by our very recent experience.
Why it’s relevant: I learnt a lot from Kurzweil and this particular essay captures both the promise and cognitive trap of exponential acceleration in a way that forces us to recalibrate how we think about the future.
If knowledge-based companies are competing in winner-take-most markets, then managing becomes redefined as a series of quests for the next technological winner—the next cash cow. The goal becomes the search for the Next Big Thing. In this milieu, management becomes not production oriented but mission oriented. Hierarchies flatten not because democracy is suddenly bestowed on the workforce or because computers can cut out much of middle management. They flatten because, to be effective, the deliverers of the next-thing-for-the-company need to be organized like commando units in small teams that report directly to the CEO or to the board. Such people need free rein. The company’s future survival depends upon them.
Why it’s relevant: I know Brian personally and his work has long been one of my go-to sources to understand the underlying dynamics of business and technology. I discussed some of this work with him on the podcast in 2019.
PLACES TO INTERVENE IN A SYSTEM
(in increasing order of effectiveness)
12. Constants, parameters, numbers (such as subsidies, taxes, standards).
11. The sizes of buffers and other stabilizing stocks, relative to their flows.
10. The structure of material stocks and flows (such as transport networks, population age structures).
9. The lengths of delays, relative to the rate of system change.
8. The strength of negative feedback loops, relative to the impacts they are trying to correct against.
7. The gain around driving positive feedback loops.
6. The structure of information flows (who does and does not have access to information).
5. The rules of the system (such as incentives, punishments, constraints).
4. The power to add, change, evolve, or self-organize system structure.
3. The goals of the system.
2. The mindset or paradigm out of which the system — its goals, structure, rules, delays, parameters — arises.
1. The power to transcend paradigms.
Why it’s relevant: Donella Meadows’ hierarchy of leverage points still remains the clearest guide for change-makers who are navigating complexity.
2025-09-08 22:21:45
Hi all,
Here’s your Monday round-up of data driving conversations this week – 10 stats, in less than 250 words.
Let’s go!
Payroll growth ↓ TS Lombard analysts warn that US payroll growth is stalling. Payroll declines have historically coincided with recessions.
A landmark AI copyright win1 ↑ Anthropic’s valuation has tripled since March, reaching $183 billion, with ~10% of the Series F raise to be paid to authors whose work was used to train their AI models.
Self-preference ↑ Job applicants using the same AI tool to write their CV as the AI system reviewing applications are 23-60% more likely to get shortlisted.
2025-09-07 10:20:12
Hi all,
Welcome to our Sunday edition, where we explore the latest developments, ideas, and questions shaping the exponential economy.
Enjoy the weekend reading!
Azeem
Given all the other places you could put your money, would you choose OpenAI? Will its return be better than buying the Nasdaq? My gut reaction was “no way”. But nothing beats pencil, paper and Excel and when I sat down with my simplistic OpenAI model and ran some scenarios, a different picture started to emerge. Here’s my take:
Two more papers came into my view this week on AI and youth employment. First is a piece of research by Harvard economists that looked at 285,000 US firms and 62 million workers to find that junior hiring among AI-adopting firms was down by 22% starting in early 2023 relative to non-adopters, while senior employment continued growing.
These findings align with ’s research using ADP payroll data, which found that workers aged 22-25 in AI-exposed occupations experienced a 6% employment decline since late 2022, while older workers in the same jobs grew 6-9%.
The second study I’d like to highlight found that since 2021, sectors exposed to LLMs have seen higher wages and more jobs, especially for young and well-educated workers. But in areas where AI directly replaces people, jobs have fallen. This, again, aligns with Erik’s paper.
I discussed the latest research in my video commentary:
Putin and Xi were caught on a hot mic musing about organ transplants and immortality. Odd talk, but biotech is starting to make the idea less fanciful. We are now developing both the atlas and the search tools to discover how human aging might be reset – and this matters hugely beyond wrinkles: aging is the single biggest risk factor for cancer, heart disease, Alzheimer’s, and most of the conditions that undermine quality of life.
2025-09-06 16:42:10
“The level of deep insight into technological trends you share is unmatched. No one is doing it like you.” – Susan, a paying member
I was discussing the OpenAI secondary share sale with a friend. Reputable sources say the firm is allowing insiders (early investors and employees) to sell up to $10.3 billion of stock at a $500 billion valuation. Apart from being great news for San Francisco realtors, it gives outsiders a chance to get a slice of this epochal company.1
Discussing this with my buddy, he asked me a question over WhatsApp:
This isn’t a question about whether OpenAI is good or even a great company. It isn’t a question about whether it will become profitable. It isn’t even a question about whether it has peaked or not. Or whether the firm has great prospects over the coming years.
The subtext of the question is this:
Given all the other places you could put your money, would you choose OpenAI? Will its return be better than buying the Nasdaq?
Let’s work through this systematically. The Nasdaq has delivered roughly 13% annual returns over the past decade, albeit with considerable volatility. For OpenAI to justify its valuation premium, it needs to outperform this benchmark substantially.
My gut reaction was “no way”. I did the mental math of the Nasdaq return and just felt that OpenAI could never compound faster than that.
But nothing beats pencil, paper and Excel and when I sat down with my simplistic OpenAI model and ran some scenarios, a different picture started to emerge.
Today, I’m going to lay out some of that thinking here.
At $500 billion, OpenAI is, today, worth roughly the same as Netflix. The streamer delivered $39 billion in revenues last year and profits before tax of nearly $10 billion. That is approximately ten times the sales of OpenAI in 2024. The AI company was, of course, deep in the red that year, losing about $5 billion.
For buyers of OpenAI’s stock today to match Nasdaq’s 13% annual return over five years, OpenAI would need to reach a $900 billion valuation.
But that wouldn’t be a good deal.
OpenAI is a privately held company with a complex shareholding structure and governance relationship. It also has a commercial deal with Microsoft which provides cloud infrastructure, access to Azure’s distribution channels and billions in capital support. This all in exchange for exclusive licensing of specific technologies and 20% of OpenAI’s revenue through 2030, up to a cap of $120 billion total return.
As a private company, it doesn’t face the same financial and governance scrutiny as a public firm. In practice, private shares with complex information rights and restricted transfer often warrant a material liquidity and governance discount. Your hurdle rate should reflect that, perhaps demanding another 10-12% for those idiosyncratic risks.
All of that means you’d need to demand a risk premium – and seek a return better than that of the Nasdaq.
In any case, while Nasdaq’s returns over the last decade have averaged 13% per annum, its three-year return has been closer to an exuberant 23%. Now, I don’t think that can be sustained. But over fifteen years, the Nasdaq has returned about 17% which seems like a decent place to start.
Applying different risk premia to that 17%, an investor taking a position in OpenAI might ask for a 20% or 25% annual return, with real implications on the firm’s five-year valuation – perhaps as high as $1.5 trillion.
At $1.5 trillion, OpenAI would sit somewhere between Tesla ($1.08 trillion), Broadcom ($1.42 trillion) and Meta ($1.85 trillion). Is this plausible?
2025-09-06 01:16:56
OpenAI is expanding a secondary share sale that lets insiders sell about $10.3 billion of stock at roughly a $500 billion valuation. That raised a simple question: is a private slice of OpenAI better than buying the Nasdaq?
I have an exclusive essay ready for members (arriving in your inboxes tomorrow!) breaking down the math and my assumptions in detail. Become a premium member to receive it in full tomorrow morning.
Ahead of this release, I did a quick live chat with EV readers to share my initial assumptions and answer your questions. We covered:
What helps OpenAI’s case,
What tempers its case,
And Q&A including on compute & energy constraints, competition and regulation.
Although I was skeptical at first, my calculations suggest that there is a path for OpenAI to deliver outsized returns. More in tomorrow's essay!
Azeem
P.S. Keep in mind that none of this is financial advice - only a thought experiment about the future.
2025-09-04 00:05:59
Hi, it’s Azeem here.
The future is being tugged in different directions by powerful forces that don’t move in unison. Sometimes they reinforce each other; sometimes they cancel out. The question is which will dominate – and how we can prepare.
Today, I invited Joe Davis, global chief economist and global head of the Investment Strategy Group at Vanguard, to share his view on this question as someone who’s studied AI and other megatrends in depth from a unique vantage point. Vanguard is one of the world’s largest investment companies, managing more than $10 trillion in assets. In this essay, Joe argues that two of the biggest forces – technology and demographics – may end up balancing each other in surprising ways.
Enjoy –
Azeem
By Joe Davis
This year marks the onset of the “silver tsunami” – the wave of baby boomers transitioning into retirement. More Americans will turn 65 in 2025 than ever before. Between now and the year 2035, approximately 16 million Americans will retire, creating profound impacts on the economy, healthcare and society at large. As the ratio of active workers to retirees falls, we could face a challenge to economic growth caused by the combined pressure of two megatrends: demographics and deficits. Fewer workers generate less output, and an aging population increases fiscal pressures by consuming more healthcare and straining Social Security.
However, history tells us that demographics are not destiny. It can be a factor, but it’s not fate. Technology (see charts below) has historically — and should again — exert a far larger influence on economic outcomes than birthrates. We need automation not as a threat to jobs, but as a catalyst for productivity. This is a central insight of my team’s research at Vanguard: AI is not merely another labor-market disruption, but a potential solution to some of the most pressing economic challenges of our time.
Imagine if baby boomers never retire. That’s the scale of productivity AI might deliver – if our models, and history, are right.
Unlike others who speculate about AI’s future, we base our views on long-term data and scenario probabilities to guide investors in assessing risk.
The product of this research, which began in earnest more than two decades ago, is the Vanguard Megatrends Model, our latest framework. This forward-looking, data-driven model provides a sense of the most likely economic scenarios through 2035, driven by an interplay of factors: technology, demographics, globalization, and fiscal debt – so-called megatrends. Behind the model is a powerful forecasting engine: a vector autoregression (VAR) that tracks 15 economic signals over 130 years of data and billions of data points. These include real GDP growth, inflation, interest rates, labor force participation, and equity valuations, among others.
In developing the model, my team looked at the future of work by examining more than 800 US occupations to analyze how AI may automate or augment each of the tasks required in these roles, rather than the roles in their entirety. We find that most American workers will feel AI’s impact – but not as a replacement.
For most jobs – likely four out of five – AI’s impact will result in a mixture of innovation and automation, and could save about 43% of the time people currently spend on their work tasks. For example, if your job includes daily tasks such as scheduling meetings, data entry, or managing projects, many of those tasks are likely to be automated. This frees up time for higher-value tasks. We find that task augmentation will enhance efficiency across more than 400 occupations, including high school teachers, pharmacists, and human resource managers.
Consider a nurse who spends up to two hours per shift manually entering data into Electronic Health Records (EHRs). With AI automation, using natural language processing (NLP), those hours can be spent on direct patient care. However, we don’t think this will result in shorter workweeks for most of us, and it won’t systematically eliminate most jobs. Instead, workers’ time will increasingly shift to higher-value-added and uniquely human tasks.
When augmentation like this occurs, the quality of a product or service goes up, boosting GDP and enhancing the value of our work (did someone say pay raise?). This mixture of augmentation and automation leads to the evolution of job roles rather than their elimination.
Yes, there are jobs with more repeatable tasks that will face decline. Computer programmers spend almost 45% of their day on programming and performance testing, tasks in which AI systems are likely to surpass human-expert level proficiency in a few years. Fortunately, many of these displaced programmers will likely transition into new AI-oriented occupations given the relative task-similarity across high-skilled computer jobs.
This productivity boost comes at a fortuitous time as baby boomers retire. If AI continues to advance and automate and augment tasks over the next 10 years, that productivity lift would put US GDP growth near 3% during the 2030s. Broadly speaking, that would be the fastest growth in the US since the late 1990s.
There are many who are skeptical of the promise of AI. Will AI follow the path of social media, which was widely adopted but did not boost productivity – and, as anyone caught zombie scrolling knows, sometimes even saps it? Or will it boost productivity only slightly, with our US economic headwinds of demographics and deficits swamping the benefits? Our model uses a probabilistic approach to predict these scenarios so that investors can prepare for both possible outcomes.
Productivity Surges scenario (45–55% probability)
Our leading scenario is that AI becomes a general-purpose technology, like electricity. This eventually lifts productivity more than the personal computer and the Internet by the first half of the 2030s. AI eventually sparks new industries, and inflation remains well in check, with deficits declining given higher tax revenues from stronger growth.
Deficits Drag Scenario (30–40% probability)
Yet, we also model that there is a concrete chance that AI disappoints – more hype than help. Meanwhile, government deficits keep climbing, interest rates and borrowing costs rise, and credit slows. Inflation is stubborn. Homeownership moves more out of reach, standards of living fall, and US growth becomes less exceptional and more similar to Europe’s growth.
Importantly, there is little that monetary policy can do to improve GDP if AI disappoints, as our research has shown that monetary policy has relatively little effect on GDP. Low interest rates in Japan, for example, haven’t stimulated growth. How individuals and organizations leverage technology to increase productivity is key to unlocking the economic growth engine.
Technological transitions are rarely smooth. But we’ve learned what not to do – ignore the worker. The workers most threatened by automation are those who perform repetitive, single-task jobs. Admittedly, governments have struggled during previous technological revolutions to adequately retrain workers for the jobs of tomorrow; the lingering social consequences of de-industrialization serve as a stark reminder. As an example, Vanguard is located just over an hour from Allentown, made famous by Billy Joel as an example of the collapse of the Rust Belt. Lost steel jobs resulted in lost wages, benefits and tax base for nearly two decades, before the steel and manufacturing industries were slowly replaced by healthcare and e-commerce-related firms (e.g., Amazon fulfillment centers).
Reducing barriers to occupational transitions, such as unnecessary educational requirements or costly certifications, provides a cost-effective means to enhance workforce mobility. From an employer perspective, the post-COVID labor market recovery highlighted the value of expanding talent pools beyond traditional channels and recognizing the portability of skills across occupations to meet hiring needs.
As my own children enter the workforce, I’ve thought a lot about the future of work and what skills they need to be successful in a technology-driven future. Human employment has always involved the acquisition of new skills as technology advances, and humans have been remarkably adaptable.
What skills can’t be easily automated? Critical thinking, creativity, emotional intelligence, and complex problem-solving are essential skills in fields where human interaction and nuanced decision-making are crucial. It’s hard to think of an occupation where this isn’t critical. Healthcare, education, and social work are areas where AI can augment human capabilities but can’t fully replace them.
And even though AI may automate many programming or IT tasks, AI itself requires critical thinking and technology proficiency to integrate it into existing workflows successfully. STEM degrees with strong analytical components may not command the same wage premium as they have in the past twenty years, but they will still be in demand.
Why do we build models to forecast future probabilities, like the future of work? Not just to guide our children’s careers or tinker with AI in the abstract, but to act. As investors saving for retirement, education, and long-term goals, we need to take informed risks and prepare for the most likely outcomes. For example, knowing the impact AI may have, we can open our aperture and consider more than the “Magnificent Seven” – the companies directly involved in AI and advanced computing – but look to which industries will benefit from AI as a general-purpose technology, what new sectors may emerge, and how these shifts could affect the economy and markets amid demographic changes, globalization, and fiscal deficits.
As our model shows, the full transformative impact of AI is likely to reverberate across industries and sectors, stressing the age-old importance of diversification. Some of these shifts will be obvious in hindsight. Most won’t be. That’s why I return to first principles: stay diversified, think long-term, and build portfolios that can adapt to acceleration. Because when transformation comes, it rarely announces itself.
Disclaimer: The views expressed in this essay are those of the authors and do not necessarily reflect the views of Exponential View or Azeem Azhar.
