|
In this issue: - Technological Progress is a Stack of S-Curves; Diffusion is a Stack of J-Curves—Long-term economic growth is generally the overlap of some static industries, some declining ones, and a few that are growing fast and diffusing benefits elsewhere. But for those high-growth sectors to really work, everything needs to be reengineered, often with an upfront cost. The extent to which people, institutions, and capital markets are willing to underwrite this temporary cost sets a speed limit on growth.
- AI Efficiency—We're automating the most annoying kinds of writing first.
- Dollar Diplomacy—The US is still offering generous foreign aid, but in a more transactional way.
- Sora—The tradeoff between distribution and respecting intellectual property rights.
- Capital Structure—OpenAI engineers a way to pay for a purchase through the stock price appreciation of a supplier.
- Reshoring Paradoxes—What can we learn from made-in-America Sharpies?
Technological Progress is a Stack of S-Curves; Diffusion is a Stack of J-Curves
There are some convenient facts about the physical world that, depending on your point of view, illustrate something about the Fermi paradox, the anthropic principle, or divine providence. We have a convenient orbit that gives us seasons, but not an annual apocalypse; surface water, but not too much; a magnetic field that protects us from solar wind; and because of an elaborate mix of evolutionary history, chemical processes at the bottom of lakes and oceans, the metabolisms of certain anaerobic bacteria, and geological processes taking place over millions of years, the planet also features a conveniently dense energy source, accessible with 19th century technology and sufficient to bootstrap a civilization to the point that it can also add fission, solar, wind, and perhaps other kinds of power to the energy mix. We're just a very lucky species.
And that luck extends to economic growth! From roughly the moment that we escaped the Malthusian trap—in the UK, that would be around when Malthus wrote his manifesto, a little earlier in the US, and a little later elsewhere—global GDP growth and US GDP growth have been remarkably stable. Even in the worst downturns, like the Great Depression, US GDP per capita was up from a generation earlier, and someone living in 1933 could buy safer food and prepare it using modern appliances. They were, on average, richer than their parents had been, even though they were much poorer than they had been a few years earlier, and even though a lot more of them were unemployed.
If you view the economy mostly through the lens of human capital, institutions, and incentives, you don't need to explain much. Every year, everyone with a job knows a little bit more about how to do their job; over time, the least-efficient operators in a given industry tend to go under, and the industry gets taken over by more efficient companies. Legal precedents accumulate, which makes the outcomes of contracts more predictable. Laws proliferate, too, but they're subject to the same kinds of selection pressure, so the bad ones (eventually) die and the good ones stick around.
But switch to a more literal view of the world, and the problem gets harder. We're always depleting some natural resources, for example—oil is technically renewable, since there's always dead plant matter sinking to the bottom of various bodies of water. But we're extracting it a few orders of magnitude faster than we renew it. In the early days of oil exploration, it was entirely possible to find ridiculous quantities of easily accessible oil—the Spindletop gusher in 1901 produced more oil than the entire industry had a decade earlier (albeit briefly). Newer discoveries don't do that, though: Guyana, whose GDP per capita chart would be impressive even if they dropped a zero, will probably produce around 1.5% of the world's oil at peak. That's a pretty big deal, especially for a country that's a little more than 0.1% of the world's population. But as far as long-term oil production goes, it's a gentle wobble on the chart.
The general story with non-renewable resources is that we find the obvious sources early, and have had to get x% more careful and creative for every x% depletion we accomplish. Or, put another way: at current levels of technology and investment, we are always doomed over the long run to spend more and more on non-renewable natural resources, and to eventually reach the point where they're the hard constraint on not just economic growth, but economic activity.
There's a coherent version of world history where we discover some key civilization-shifting resource, like coal or oil. And then we exploit all the easily-accessible kinds, don't have the resources to keep going, and we revert back to the prior standard of living. And because that standard of living prices more people out of literacy, and makes things like maintaining accurate written records in the first place an expensive luxury, the lived experience of people in that post-collapse order is that every so often, they encounter some miraculous artifact that could only have been created by supernatural means. (An 8th or 9th century English poem refers to Roman architecture as "the work of giants." Poetic license, sure, but it works because the poet is accurately contrasting what his contemporaries could build with what had been built before.)
And if it's not bad enough to always be just a few generations away from Canticle for Leibowitz territory purely because hydrocarbons burn, metals rust, and everything wears out with use—that Old English poet was writing in the aftermath of a collapse in the social order, which broadly reduced standards of living among basically every class that wasn't already on the edge of subsistence, and which, for that last class probably killed quite a few of them. (Since the collapsed social order in question was partly devoted to extracting as much of their economic output as possible, some of the survivors did just fine.) If we start to lose social trust—if you're not sure that purchasing a product online means you'll receive it, and the seller isn't sure that they'll actually collect the money from you, or if everyone building a datacenter knows that some fraction of the budget will obviously be embezzled, this drives a deadweight-cost wedge between every transaction, and means that many mutually-beneficial ones don't happen. And if the upside from that cheating actually makes the cheaters better-off, then the equilibrium level of trust lurches downwards. These things are cyclical; it was a big mistake for Nixon to let the Watergate breakin and cover-up happen when the Weather Underground was campaigning for him so successfully. But this has the feel of a martingale bet, where a system at rough equilibrium wobbles out of it, and corrects, but there's no logical limit to the degree of wobbling and no law of nature that requires it to wobble back.
What gets us out of this is that technological progress is a stack of S-curves, one on top of the other. Industries grow, and mature; specific work processes reach some asymptote where they've either reached the theoretical limit of their efficiency or have been overtaken by something else. We find substitutions everywhere, whether that's switching from human- and animal-powered transportation to cars, shifting from personal accumulation of material goods to personal accumulation of financial claims backed by the ability to produce goods, using on-device translation tools instead of learning a foreign language or hiring a personal translator, or even more abstract ones like replacing warehouses with better supply-chain management.
But these processes, too, reach a limit. And, more to the point, they start with one: the way we experience modern, ridiculously complex supply chains—taking a flight, taking a Waymo, asking an LLM a question, downloading an app—all rely on layers of infrastructure, much of which was built with other systems in mind. All of these technologies are born as strictly worse replacements for whatever came before; a streaming video app is a TV that you can carry anywhere that you want to watch a limited selection of low-resolution videos with tinny audio. (Also, unless you were watching a very special video that rewarded patience, you probably wanted to be on WiFi.) A car was mostly recreational when cities were designed around pedestrians, mass transit, and horses. A computerized spreadsheet isn't necessarily faster than a physical one if the user doesn't know how to type.
Eventually, all of these problems get resolved, and processes get built around whatever the dominant technology is. Some of this is subtle; one sign of the dominance of phones is that videos are shot vertically by default—i.e. meant for consumption on phones—rather than horizontally, for viewing on different screens. (There were two coup attempts in Turkey in 2016; the failed one against Erdogan, and the successful one against horizontal video, when Erdogan addressed the country in vertical-video format.) Some of these tech-infrastructure feedback loops are gargantuan; whole industries exist because human-operated cars are such a common default, and city layouts and company structures will probably reflect this long after it's changed. A business like Kroger or Costco only makes sense in a world where there are ubiquitous personal vehicles with ample cargo space, but these businesses will probably exist in some form even if drone delivery kills off a substantial chunk of non-experiential physical world shopping.
It's very rare for a new technology to produce an immediate productivity improvement for everyone, but it's also very common for it to produce obvious wins in a narrow set of uses and then to expand from there as costs decline. So the arrival of some S-curve, where we as a society learn how to cope with coal, or railroads, or customer loyalty programs, or structured products, or the like button. And then a surprising number of economic relationships get built around them, and they become an invisible part of the social fabric.
What this requires is some kind of permission structure for underwriting J-curves, where the initial result of using some new process is a steep drop in output, but the longer-term result is a new, higher ceiling. On an individual level, that's sometimes tolerable—you can just decide that whatever your peers are doing in Excel, you'll be doing in Pandas or Polars or something. But one of the reasons these J-curves matter is that many technologies depend on complements; we'll just live in a different world if the default way to share a financial model switches from opening an attachment to cloning a Github repo, even if the underlying information is exactly the same.
But that kind of change does, eventually, happen. There was a time, in living memory, when the default way to share financial information wasn't sending an Excel file, but once it started to be the way people thought, it was hard to resist. In other words, the low point of the J-curve got shallower as adoption grew. That's part of what AI companies are pushing right now: the more CEOs there are who are sending memoranda on how "use AI" is the default solution to most business problems, the more that becomes the default solution, and the more we'll assume that interpersonal interactions in a professional context are mediated by two layers of AI unless they happen in real time and in-person. The more that's the default—the more we think of intelligence as something you access via API rather than by buying it for forty hours a week at a time—the easier it will be for the next institution to adopt that as a default. And that's the track that it takes to add another S-curve to the stack of S-curves that constitutes surprisingly stable indefinite economic growth.
Diff JobsCompanies in the Diff network are actively looking for talent. See a sampling of current open roles below: - A hyper-growth startup that’s turning the fastest growing unicorns’ sales and marketing data into revenue (driven $XXXM incremental customer revenue the last year alone) is looking for a senior/staff-level software engineer with a track record of building large, performant distributed systems and owning customer delivery at high velocity. Experience with AI agents, orchestration frameworks, and contributing to open source AI a plus. (NYC)
- Well funded, Ex-Stripe founders are building the agentic back-office automation platform that turns business processes into self-directed, self-improving workflows which know when to ask humans for input. They are initially focused on making ERP workflows (invoice management, accounting, financial close, etc.) in the enterprise more accurate/complete and are looking for FDEs and Platform Engineers. If you enjoy working with the C-suite at some of the largest enterprises to drive operational efficiency with AI and have 3+ YOE as a SWE, this is for you. (Remote)
- A leading AI transformation & PE investment firm (think private equity meets Palantir) that’s been focused on investing in and transforming businesses with AI long before ChatGPT (100+ successful portfolio company AI transformations since 2019) is hiring Associates, VPs, and Principals to lead AI transformations at portfolio companies starting from investment underwriting through AI deployment. If you’re a generalist with deal/client-facing experience in top-tier consulting, product management, PE, IB, etc. and a technical degree (e.g., CS/EE/Engineering/Math) or comparable experience this is for you. (Remote)
- A transformative company that’s bringing AI-powered, personalized education to a billion+ students is looking for elite, AI-native generalists to build and scale the operational systems that will enable 100 schools next year and a 1000 schools the year after that. If you want to design and deploy AI-first operational systems that eliminate manual effort, compress complexity, and drive scalable execution, please reach out. Experience in product, operational, or commercially-oriented roles in the software industry preferred. (Remote)
- YC-backed founder building the travel-agent for frequent-flyers that actually works is looking for a senior engineer to join as CTO. If you have shipped real, working applications and are passionate about using LLMs to solve for the nuanced, idiosyncratic travel preferences that current search tools can't handle, please reach out. (SF)
Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up. If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority. Elsewhere
AI Efficiency
A new paper finds that LLM-generated writing shows up in about a quarter of corporate press releases, as well as a small but still significant fraction of other official documents. As someone who's helped write actual press releases in the past, this doesn't seem like a negative at all: part of the reason they exist is that the headline looks weird without any associated content, and a terse press release sends its own kind of message. So there's some amount of make-work around making sure the people who read the press release consider it business-as-usual, and a bland ghostwritten quote for any of the principals involved in whatever is being released (if it were an interesting statement, they'd probably save it for an interview). The press release is like a sitcom, in that it's a medium whose content has to be shoehorned into a standard form, and it's similarly forgettable and worth automating away.
Dollar Diplomacy
The Economist calls attention to the fact that the Trump administration is happy to provide generous foreign aid, but mostly to countries that are aligned with the US agenda ($). In a way, this is a form of belt-tightening. A rich country that doesn't need to count every dollar can be generous and argue that, in some hazy sense, this generosity will rebound in the form of goodwill in previously hostile or unaligned nations. But a country that's paying closer attention to its expenditures will probably want to focus on either flipping toss-up countries towards the US sphere of influence, or helping countries whose current leader is pro-US remain in power a bit longer. To an extent, it subverts Argentinian democracy for the US to support the country in order to keep a particular politician in power. But there's also some hard-headed realism here: part of what voters vote for is perception from other countries, and the US's perception matters a lot.
Sora
OpenAI chose to release its most recent video model, Sora, as part of a social feed product—you can create videos and immediately share them. As with many early social networks, a lot of the content is self-referential—Sam Altman has lots of cameos—but it's slowly broadening from there. More interestingly, Sam Altman says one of the big open questions is about intellectual property. There's generally a price at which some celebrity or character can show up in an ad, and often no market-clearing price at which they can be placed in some truly offensive scenario that anyone attempts to monetize. If AI companies start to control more distribution, then there's a good chance that this will replicate the existing status quo: you can buy lots of knockoff apparel featuring billion-dollar characters like Spongebob and Bart Simpson engaged in behaviors that give Paramount and Disney's legal teams a heart attack. But for the best production values and best distribution, you'll need those characters to follow the brands' rules.
Capital Structure
OpenAI and AMD signed yet another big, roundtrip-flavored chip deal this morning. OpenAI will buy AMD chips, and will get warrants to buy AMD shares in exchange, potentially taking up to a 10% stake. One of the ways to understand a cycle is to ask where the funding is coming from. Typically, in the earliest stages, the funding source is basically opportunity cost: people quitting their jobs building desktop applications to make websites instead, or ML engineers taking some time off between jobs to really understand those new papers on the transformer architecture. Then, as capital needs go up, equity financing gets bigger, and, when equity markets aren't abundant enough and when there's a consensus that even the low-end outcome is pretty good, things shift to debt. (After the overshoot, the main source of funding shifts to the operating cash flows of the survivors, but we're not there just yet.) OpenAI's deal is partly set up so that the funding source is the public equity market, which makes OpenAI's AMD warrants worth more, so they can realize those gains and pay for AMD's chips. The negative view is that AMD using its own equity to collateralize a joint venture that will produce earnings for it is exactly what Enron did, but the positive view is that there's an old equilibrium where AMD is worth a bit less, a new one OpenAI can make happen where they're worth more, and warrants are AMD's way of paying OpenAI to shift reality in a direction that helps them.
Reshoring Paradoxes
The WSJ has a story about how Newell reshored production of Sharpies, and has been able to ramp up efficiency and wages without increasing headcount ($, WSJ). One of the secret ingredients is that they're making Sharpies, a brand name literally synonymous with the category, so they can underwrite some capital expenditures based on having confidence that they'll still be selling the product, and selling it at a premium, over the useful life of those investments. At the same time, a factory located in the US for a fairly low-value product also means that they can be more responsive to market trends, without the multi-week lag of ocean freight. That's a good template for insourcing: take something where the brand name is a constant, but the specifics always vary. (This is one reason the US auto industry has stuck around; a Ford today is something different from a Ford a decade ago, so flexibility and shipping speed matter. But Ford still means Ford.)
|
