Welcome to the 784 newly Not Boring people who have joined us since last week! If you haven’t subscribed, join 233,971 smart, curious folks by subscribing here:

Today’s Not Boring is brought to you by… Speakeasy

As the old saying goes, “Get someone to use your product, get a user for a day; get someone to integrate your API, get a customer for life.” Or something like that. 

I’ve been a fan of the API-first business model since writing about it back in 2020. Half the battle is building APIs that handle mission critical but non-core work for your customers. The other half is making it as easy for them to integrate your APIs as possible. Speakeasy can help integrate API users 50% faster. 

Creating a frictionless API experience for your partners and customers no longer requires an army of engineers. Speakeasy’s platform makes crafting type-safe, idiomatic client libraries for enterprise APIs easy. That means you can unlock API revenue while keeping your team focused on what matters most: shipping new products. Make SDK generation part of your API’s CI/CD and distribute libraries that users love at a fraction of the cost of maintaining them in-house. Let Speakeasy handle your mission-critical, non-core work so you can do the same for your customers. 

If you know what any of those terms meant, and if those pain points sound familiar, you should check out Speakeasy.

Try Speakeasy for Free

Hi friends 👋,

Happy Wednesday! I’m sending a day late because I spent the weekend at the Roots of Progress Institute’s Progress Conference and the last couple of days in San Francisco.

As much of a New Yorker as I am, I gotta say, it feels like SF is back. People are turning ideas into products, and at the right price point, products turn into progress. This is an essay based on conversations from the past week on how to do that.

Let’s get to it.

What Do You Do With An Idea?

I’ve noticed a common refrain speaking with founders building physical things: 

“This is an old idea, actually, from the…” 

I interrupt, a shit-eating, Cheshire grin spreading across my face: “1950s or 1960s, right?” 

Right, they say, adding some specific variant like, “It’s from a 1958 paper.”  

But the paper was obscure, or Soviet, or the idea wasn’t technically possible or economically feasible with the tools of the day. So it collected dust, forgotten and waiting to be rediscovered. 

Good ideas aren’t getting harder to find. We just need to use the ones we have. 

Are Ideas Getting Harder to Find? 

I spent the weekend at Progress Conference 2024, hosted by Jason Crawford, Heike Larson, and the team at the Roots of Progress Institute. It was awesome. A group of my favorite progress thinkers and doers from the internet came to Lighthaven in Berkeley to discuss the question: how do we make more progress? 

There was a range of talks and topics, but the topic that came up the most was some version of a debate over the 2020 paper by Nicholas Bloom, Chad Jones, John Van Reenen, and Michael Webb, Are Ideas Getting Harder to Find?

If ideas are getting harder to find and if the “low hanging fruit has been picked” – and the authors argue that they are and it has – then it’s unsurprising that we are stagnating. 

Economic growth, they write, is a function of ideas, which is a function of research productivity and number of researchers. 

To determine whether new ideas had indeed become harder to find, they looked at three specific case studies – Moore’s Law, seed yields, and medical innovations – and at the growth of the economy as a whole. 

They found that in all the cases they examined, it’s taken more researchers just to find the same number of ideas over time, which means that ideas are getting harder to find, which implies that we either need to throw an exponentially increasing number of researchers at problems just to maintain our current growth rate (maybe possible with AI!) or settle in for a long stagnation. 

I am not an economist, so maybe this is just how these things work, but the whole argument strikes me as incredibly circular: 

• The paper defines ideas as things that lead to Total Factor Productivity (TFP) growth.

• TFP is the portion of economic output growth that cannot be explained by increases in the quantity of inputs (like labor and capital). 

• The paper then measures research productivity as TFP growth divided by research input.

• Finally, it concludes that because this ratio is declining, ideas are getting harder to find.

By defining ideas in terms of TFP growth, the paper essentially assumes its conclusion. 

The paper does provide valuable data on the relationship between research inputs and certain economic outputs over time, and it finds this pattern across various domains, which suggests there's something meaningful happening. And the authors have forgotten more about economics than I’ll ever learn, so maybe I’m missing something. 

The problem is, the paper might actually lead to the wrong conclusion: that we need to throw more researchers at the problem, that we just need to produce more ideas. 

For one thing, it might suggest that now that AI is here, we’re good. While the number of human researchers won’t be able to scale exponentially to offset declining research productivity, maybe the number of AI researchers can. If you assume that ideas are the bottleneck, then maybe you just hold out for a sufficiently intelligent model. 

The question of whether AI would just solve progress came up multiple times at the conference. 

The most common response was: no, ideas are not the bottleneck. 

On the opposite end of the spectrum, it might suggest that we are really and truly screwed. Total Factor Productivity really has fallen off since the 1970s, and if ideas just keep getting harder to find, and ideas are what drive TFP, then we might just be in for greater and greater stagnation.

Both of these are dangerous but passive. Wait and see. 

A third option, implied but not recommended by the paper, would be to throw more at research: more people, more education, more money. If ideas are getting harder to find – if it takes more “effective researchers” to generate the same number of ideas – then we need to produce more researchers to at least match the rate of decline in research productivity. 

That strikes me as equivalent to observing that a football team’s yards per carry is declining and deciding that they need to run the ball more to make up for it. To torture this analogy a bit, it’s maybe even more equivalent to signing more running backs, whoever’s available off waivers, and then running the ball more with them, instead of fixing your offensive line. Worse, there are only 53 roster spots, so every new running back you sign means cutting someone who could be doing the blocking and tackling!

We will always need new ideas. In this house, we subscribe to the David Deutsch Principle of Optimism: “All evils are caused by insufficient knowledge.” 

And to be fair, the paper isn’t trying to suggest that a lack of ideas is solely responsible for the slowdown in TFP growth. It’s simply arguing that the assumption of constant research productivity doesn’t hold up, and it makes its point. 

But the simplistic takeaways – that ideas are getting harder to come by so either we’re doomed, can expect AI to save us, or need to throw more researchers at the problem – are dangerous and, I think, wrong. 

There are plenty of good ideas waiting to be turned into progress. There’s an idea backlog. 

The Idea Backlog

So why are so many companies being built today using ideas from the 1950s and 1960s? 

Let’s look at this chart again: 

One thing that the paper omits is any discussion of why, if research productivity had been declining fairly steadily and the number of effective researchers had been rising fairly steadily since the 1930s, there was such a dramatic fall-off in TFP starting in 1973. 

That is a major omission! If the slowdown in TFP growth can be explained by ideas getting harder to find, why did growth hit a brick wall in 1973 as opposed to slowly decaying as soon as ideas became harder to find? 

Well, one question to ask might be “Which factors contributed to TFP growth in the last good decades, the 1950s and 1960s, and when were the ideas for those factors discovered?” 

I asked that question, to both Claude and Perplexity, and both cited: Petrochemicals/Plastics, Electronics / Semiconductors, Aviation, Antibiotics / Medical Advances, Mass Production Techniques, all of which were conceived of in the 1920s through 1940s or earlier! 

By the time those ideas’ productivity was realized, they were relatively old ideas. As Perplexity concluded, “In conclusion, while the 1950s and 1960s saw remarkable TFP growth, this ‘golden age’ was largely built on technological innovations and research from earlier decades, particularly the 1930s and 1940s.” There was a two decade lag. 

So given the TFP fall-off in 1973, it’s not surprising that so many ideas from the 1950s and 1960s are still available to entrepreneurs today. It’s as if they were frozen in amber. 

The question isn’t whether good ideas have gotten harder to find, but why it got so hard to use the good ideas we already generate. 

In one of the best talks of the conference, Eli Dourado argued that ideas aren’t getting harder to find; they’re getting harder to use, thanks to the three rules put on innovation:

This is the precautionary principle ad absurdum. The Nuclear Regulatory Commission (NRC) is an obvious punching bag, and one I’ve punched many times in these pages. The challenge isn’t just that they make it hard and expensive to license new reactors, but that they make it illegal to experiment and iterate on new reactor designs. 

Used to be, a man was allowed to fuck around and find out. That’s how progress happened. Edison’s electrical wires electrocuted people. The Wright Brothers faceplanted into the Kitty Hawk sand during test flights. The boilers on Vanderbilt’s steamships occasionally exploded. Then they iterated, improved, even undertook new research to create new and improved ideas to address the obstacles that can only be exposed by interaction with the real world. 

Then, Americans’ attitude towards risk shifted. I wrote about this in Riskophilia. For that post, I asked ChatGPT to make me a chart of America’s attitude towards risk by decade, and while certainly unscientific, there is a stark difference between the 1770s - 1960s trendline, and the 1960s to 2020s trendline. The timing lines up well with the slowdown in TFP growth. 

Risk aversion certainly isn’t the only factor. Figuring out WTF happened in 1971? is an ongoing debate that has many answers, from declining per capita energy usage to the move off the gold standard to the shift to innovation in the digital world. As Nobel Prize-winning economist Robert Solow observed in 1987, “You can see the computer age everywhere but in the productivity statistics.” This is Solow’s Paradox. 

The point of this essay isn’t to relitigate the causes of the slowdown, but to suggest that it has very little to do with our inability to discover good ideas. We can’t even make use of the good ideas that have been patiently waiting for us for over half a century! 

At the end of his talk, Eli suggested a recipe for growth: 

Figure out something you want to build that is technically and economically feasible, figure out what’s standing in the way, and then use all of the resources you can bring to bear to obliterate those obstacles. 

I love it and I think we should do that. 

I also think there might be something else at play, because people are reanimating old ideas, despite the obstacles, more frequently than they have over the past few decades. 

Why are people building ideas again?  

Here’s another question: if ideas are getting harder to find, and regulation has only gotten worse, why are so many people building things now? 

There are two very interesting points, between which, for the sake of argument, I’m going to draw a line. 

First is the observation I’ve made throughout: that people are building companies based on ideas from the 1950s and 1960s. 

This is a very real thing. Earlier this week, I met with Tyler Hayes at Atom Limbs to see the robotic prosthetic he and his team are building. After he slipped the cuff on my arm and as we were waiting for the system to boot up, he asked what I was working on. I told him about this essay, and the ideas from the 1950s and 1960s idea. He laughed. “A lot of our work,” he said, “is based on this book from 1967. 

From Idea to Arm in 57 Years

Here’s another one. Jason Carman recently did an S3 on Longshot Space, which is building the world’s biggest gun to shoot things into space.  

It’s an awesome idea, and an old one! Canadian ballistics expert Gerald Bull came up with the idea to use a gun to shoot things into space in the 1950s, and Project HARP, which he inspired and ran as Lead Scientist, actually fired more than 100 missiles into the ionosphere.

Project HARP was funded by the Canadian and US militaries, who shut the project down in the late 1960s due to opposition from critics, “growing bureaucratic pressures,” financial and political pressures from the Vietnam War, and NASA’s focus on traditional rockets. After he lost funding, he did artillery projects for a number of countries, and was assassinated in 1990 due to his alleged work developing a supergun for Saddam Hussein. 

I could go on. This is a surprisingly common pattern. But the show must go on, so ask your favorite deep tech or hard tech founder if their idea is based on something from the 1950s or 1960s and watch what happens.

Second is the recent resurgence in nuclear power driven by the tech giants’ demand for huge amounts of clean, reliable energy to power their AI data centers. 

Nuclear power is an old idea, from the 1930s and 1940s, and one that America implemented rapidly through the 1960s and 1970s, when it seemed as if the demand for electricity would continue to grow apace with the Henry Adams Curve. It didn’t. 

There are a lot of reasons that nuclear power fell off. The NRC is certainly partially to blame – at the very least, they’ve made it much more difficult and expensive to build and iterate on nuclear technology. But one important reason that nuclear fell off is that the demand just wasn’t there. 

E2: The Untold History of Nuclear Energy Packy McCormick | Turpentine Episode

Now, the demand is back. Microsoft is restarting Three Mile Island. Amazon is investing in Small Modular Reactors. Google is buying up to 500 MW of power from Kairos Power. 

In a talk at Progress Conference, Julia DeWahl called this demand – from both Big Tech and the DoD – “premium power.” In other words, these buyers are willing to buy power more expensive than they could get from other sources in exchange for capabilities that only nuclear can provide. In Big Tech’s case, that means 24/7, reliable, clean power. In the DoD’s case, that means reliable power that is easy to transport. 

This premium demand is often critical to the development and diffusion of technologies. Certain buyers can support a technology when it’s more expensive than alternatives in the market, which provides an initial push down the cost curve. With experience, manufacturers can drive down costs to the point that they’re accessible to the broader market. 

As two examples, the military and NASA served as premium demand for both solar power and semiconductors in the late 1950s and early 1960s. 

In 1958, Vanguard I was the first satellite to use solar cells. They paid $1,000 per watt. Today, a watt of solar comes in at less than $1. 

In the late 1950s and early 1960s, the military bought nearly all integrated circuit production for use in the Minuteman missile program, and NASA’s Apollo program bought thousands of integrated circuits. Then, Moore’s Law took over. 

Then, starting in the late 1960s, both NASA and the DoD’s budgets fell off. 

Here is the DoD budget as a percentage of GDP since 1960: 

And here is NASA’s budget as a percentage of the Federal budget since 1958: 

These drops happened around the same time that America became more risk-averse and the same time that we stopped growing our per capita energy usage. At the same time that it became harder and more expensive to test and iterate, two important premium buyers dramatically cut spending. 

Taken together, these factors seem to explain why TFP fell off in 1973 more than the idea that ideas have gotten harder to find. There are more, but these make the point. 

Which brings us back to the question: why are people building things again now? 

Here’s my hypothesis: 

Ideas have gotten more expensive not to discover, but to build, at the same time that demand for premium versions of things slowed their spending. That meant that the initial push down the cost curve never happened, which meant that ideas remained more expensive to implement, which means they never got to the normal part of the demand curve. 

Now, two things are happening. 

1. Technology is making old and new ideas cheaper and easier to build. 

2. There’s a reemergence of “premium demand.” 

Ideas Need Adoption to Make an Impact

Ideas alone are not enough. They need to be techno-economically viable, buildable for a low enough cost that the market is able to adopt them. There is no progress without adoption. 

Premium demand can help technologies that have a path to market viability reach that point. It’s a bridge. It’s not charity, or even grants. These products offer capabilities that no other products do and for which the government and Big Tech are willing to pay higher prices to access those capabilities. 

Big Tech is buying nuclear power out of its own selfish interests, and the AI it produces may show up in TFP growth, but nuclear power won’t have a large direct impact on TFP while it sits behind the meter powering data centers. That will come as a result of building and manufacturing more reactors and bringing them down the learning curves. Until then, it has powerful, deep-pocketed, and motivated allies to help fight the costly hurdles in its way. 

The same can be said for capabilities that the DoD might buy – whether drones or supersonic planes – which will show up first in the military context and diffuse to consumers as companies scale up manufacturing and bring down costs. NASA and the DoD were the premium demand for SpaceX’s early launch; now, the benefits can be felt by anyone who uses Starlink to access the internet. 

For a certain set of technologies, this bridge is critical. But something else is happening, too. 

Despite the slowdown, technological progress has marched on. Compared to 1973, entrepreneurs have a cornucopia of lower-cost, higher-performance technologies at their disposal with which to build. AI, CRISPR, batteries, solar, software, launch, materials, sensors, and an ever-expanding list of inputs continue to improve and get cheaper. 

Atom Limbs’ AI prosthetic is possible, at a price the market will pay, because of advances in sensors, actuators, batteries, and machine learning. 

Longshot Space uses advanced control systems to inject gas directly behind the projectile as it moves down the barrel at sub-millisecond precision, a capability that Gerald Bull could only have dreamed of. Plus, the DoD is a potential early customer for Longshot’s hypersonic capabilities, which would help fund the development of the ultimate mission of launching mass into space more cheaply and helping humans, and our progress, scale beyond earth. 

Almost any 1950s or 1960s idea that’s being brought back to life today benefits from dramatic cost and performance improvements compared to the technology available at the time. 

Perhaps what’s happening is that those cost declines have outrun the cost increases from regulation and other countervailing forces put up over the past fifty years. More simply: it all comes down to economics. The factors for and against progress can be boiled down to the all-in costs to turn ideas into scaled products, and whether there are enough buyers willing to pay more than that cost. 

A counterpoint might be that there’s a lot of economic interest behind scaling Moore’s Law or improving crop yields, and yet, new impactful ideas in those fields are getting harder to find. My bet would be that while those particular tech trees are reaching their end, and exhibiting diminishing returns to researchers, new tech trees, driven by lower input costs and premium demand, will take over to move progress forward. That certainly seems to be happening with GPUs. 

I’m not an economist, but I would love to see somebody analyze this.

If my hypothesis is true, it aligns well with my belief that we’re entering a new Techno-Economic Paradigm in the Carlota Perez sense, a Techno-Industrial Revolution, and that TFP growth may re-accelerate.

What Do You Do With an Idea?

One of the joys of parenthood is that you get to read a lot of books with your kids. One of my favorite books to read with Dev and Maya is Kobi Yamada’s What Do You Do With An Idea?

I don’t want to spoil it, but after feeling a little embarrassed by his idea, trying to hide it away and protect it, what the protagonist realizes is that what you do with an idea is… you change the world. 

Ideas are the lifeblood of progress, but they only work when you can actually use them in the world. They don’t work tucked away in brains or even published in research papers. 

Turning ideas into progress is a complicated thing, influenced by culture, regulation, economics, technology, and the efforts of individuals. 

Maybe, for fifty years, turning ideas into progress has been more expensive than it’s worth. And anyway, driven by software, GDP has continued to grow apace, even if the unexplainable part of it has not. 

US GDP Per Capita Since 1960

Constant 2015 $

Maybe there wasn’t the urgency to do new hard things when a combination of the old hard things, software, and capital markets kept GDP growth on-trend. 

For whatever combination of reasons, that seems to be changing, and we might be finding ourselves in this beautiful place, armed with new premium demand and better technology with which to unblock a tremendous backlog of old ideas that previous generations have left on the shelf. 

Instead of harder-to-find ideas, we might have an abundance of great ideas for those with the skill and entrepreneurial vigor to bring to life. 

That would imply that we don’t necessarily need more researchers; we just need more people putting the ideas the researchers come up with to work. 

In addition to making new Einsteins  - genius researchers will always be important – we need to make new Vanderbilts and Fords and Wrights. Tinkerers who can pluck from the pile of withered ideas and combine them in new ways. We might also need new Librarians, who can help surface those great old ideas, and track which technologies and costs might make them economically viable after all of these years. Maybe that’s where AI can be most helpful. 

Of course, we need to reverse the regulatory ratchet that makes new ideas and old ones more expensive to build. We need more people working on policy that helps drive down the cost of implementing good ideas. The faster and less expensive it is to build, test, and iterate, the smaller the gap between a good idea and its contribution to progress. 

And as we start to turn more ideas into more products, I suspect that new ideas will become surprisingly easy to find. Just as models need new training data to get smarter, we need to try more things to discover which questions our researchers should ask next. 

Thanks to Claude for editing, to Eli for letting me use his slides, and to Jason and Heike for putting on an incredible conference.

That’s all for today. We’ll be back in your inbox on Friday with a Weekly Dose.

Thanks for reading,

Packy