What Part of the System Needs to be Smart?
One debate that starts out technical, but devolves into an ideological one starts with this question: for a given network of computers, what is the most ideal way to distribute their power? The answer to this question has two extremes, one of which is a fairly non-hierarchical network where most participants are fully-fledged devices that compute locally, store their own data, and interact as peers. At the other end of the spectrum, thin clients aren’t much more than a monitor, peripherals, and a connection to some beefier server somewhere far away.
The tradeoff between distributed and centralized computing power can sometimes be framed in terms of limitations on bandwidth, latency, utilization rates over time, etc. But considering that tradeoff also pokes at deeper questions about the structure of computing and the economy at large.
One argument in favor of the broad category of thin clients is that there's a tradeoff between simple interfaces and powerful tools. A product like Uber or Doordash is partly a way to take a hugely complicated question—how do we optimize a logistics network to quickly and efficiently deliver people or goods on demand, and how do we tradeoff between speed and certainty—and turn it into a single button push.
That's an impressive feat! And sometimes increasing the sophistication of the underlying system means decreasing the complexity of the interface. Since modeling errors compound over time, and since traffic jams are not linear (one additional car has either roughly zero effect on average speed or leads to a traffic jam), it's much harder to promise an arrival time at a destination than it is to promise a vehicle as soon as possible. But a decent fraction of rideshares are trying to get somewhere at a specific time with as little waste as possible. A more full-featured local system with lots of options and flowcharts is one way to achieve this, and another is to handle all of that complexity on the backend and just ask the user what they want.1
Thin clients can get increasingly powerful—in terms of what they do, not in terms of how much control users have—as whatever backend they access changes. You've never controlled your Facebook or Instagram account, in the sense that it's always somebody else storing the data and deciding how to display it. But over time, those services have gotten far more sophisticated in how they display content. What you get when you type words into Google has also changed a lot over time; the system has gone from being entirely dependent on relationships between texts to being so smart that power users are the ones who are good at telling the search engine to be dumb, by, for example, using quotes to suppress Google's aggressive search for synonyms, or using the date function to tell Google that you're looking for historical instances of something that just happened again.
The question of where complexity should reside ends up being a question about what computers really are. A fat client approach is that a computer is something close to a chief of staff, helping a decisionmaker execute their ideas and keep track of everything they need to know. The thin client model is that a computer is a simplified interface on some elaborate how-we-manage-things-here model, with a flowchart for every situation leading to one and only one answer. It’s a question of whether the user should be computing or should just be inputting some simple data and getting an answer.
The thin client model is prolific because it’s scalable, in two senses:
- Traversing a big, interconnected graph by hopping from one node to the next is a slow process, even if the hops are generally in the right direction, and this kind of decentralization creates lots of overhead, and
- There are literal cost benefits to buying storage space by the exabyte, or to having enough memory for global peak demand rather than for any device's peak demand.
But there are downsides. The ownership question is a big one. Increasingly, companies and individuals own access to their data but don't own the data itself. This is fine the vast majority of the time, but turns out to be annoying when there's some opaque dependency that's utterly unfixable, like a hacked Gmail account or an AWS outage, that makes the rest of a system stop working. The critical infrastructure layers do have a strong incentive to keep themselves running, but it's not fully aligned with the incentives of their users.
And the thin-versus-fat client phenomenon is much, much broader than the question of whether your social media profile should have a local backup and fully portable friend graph by default. It also applies to economic actors who can treat some layers that they interact with as fully trusted and completely capable of handling complex operational details:
- There has been a long trend towards US retail acting as a frontend for China's manufacturers. This was mutually beneficial, at first because of China's extremely low wages, and over time because earnings from exports were reinvested in infrastructure. Eventually, Covid seemed to do what we thought trade war concerns were going to, and led more companies to diversify supply chains. This made companies that access that part of the supply chain implicitly closer to a full-stack model than to a thin client: they still didn’t own their manufacturing, but instead of plugging into one existing system, they were sourcing from many places at once and carefully designing a resilient supply chain.
- Merchants on Amazon are increasingly encouraged to avoid thinking about demand generation, shipping, and storage. Every one of those can be solved by giving Amazon money, albeit not necessarily on the most favorable possible terms.
- The chip industry, by necessity, needs to abstract some complexity away. The supply chains are too unwieldy to be owned by a single company: too many highly-specialized devices whose components are also extremely specialized. But that's creates multiple chokepoints.
- Russia's long record of uninterrupted natural gas exports to Europe, even when Cold War tensions were high, made cheap natural gas an unquestioned input into the European economic model.
- Interacting with and relying on the dollar system gave countries more access to trade and finance, but also meant that they were more vulnerable to US sanctions.
Decentralized computing, deglobalization, and full-stack companies are really three instances of the same general trend. More centralized systems are powerful, but they're also brittle. Decentralized systems are expensive, but they're resilient; a world with a more diversified and local energy system, more spread-out manufacturing, and more peer-to-peer networks is less dependent on a few systemically important actors.
Controlling one’s own destiny is complicated, expensive, and risky. When companies describe their strategic decisions in these terms, they’re usually justifying a big and unpopular investment push. And many of these pushes don’t work out. Division of labor is a powerful concept, and the structure of supply chains as well as the structure of computer networks reflects the fact that redundancy has a cost and some decisions work best when they’re repeated. But over time, this model reduces agency, and discourages careful analysis of why the network or supply chain is structured the way that it is. Eventually, someone who relies on distant abstractions can find that their economic niche has disappeared for reasons that are completely inscrutable to them.
The shift doesn't have to be binary, though. One way it can happen is for some participants in the stack to accept that they're truly neutral service providers, and that moving up or down the supply chain represents a costly strategic sacrifice even if it's temporarily profitable. For companies, the right approach will be to decide early on if their customers should view them as a utility, or as an eventual competitor.
Disclosure: Long AMZN, long a little META as well. (Thesis there is that their current margin compression is voluntary, but that the most cost-effective way to hire people to build the metaverse is to ensure that their equity comp is worth something.)
A Word From Our Sponsors
Tegus is the first port of call for M&A professionals and institutional investors ramping up on an industry or company.
Get access to a database of 35,000+ expert call transcripts, spanning 5+ years, or schedule expert calls through the platform for a fraction of the usual cost.
When thousands of research analysts are pooling their expert calls into an on-demand database, using Tegus is table stakes. It's the leading platform for due diligence and primary research.
See the power of a Tegus subscription, and get up to data parity with your competitors, with a two week free trial through the Diff.
Gay dating app Grindr went went public via SPAC, and rose as much as 515% from their prior close before dropping to a mere 214% gain. A fun feature of SPAC deals is that this first-day response means exactly the opposite of what it would mean for a traditional IPO. With an IPO, if shares rise rapidly on day one, it's a deal lots of people wanted to get in on, pushing the price up. But since shares of the SPAC trade ahead of the deal, and since shareholders can choose to redeem their shares for cash when a deal closes, the supply of shares goes down when investors are more pessimistic.
Which is normally fine, and actually a healthy feature for an offering; there's always uncertainty with a conventional IPO around how many shares could have been sold, and some companies regret how little they sold when their offering pops.
But it also means that a sufficiently unpopular SPAC deal is a great recipe for a short squeeze. In the Grindr SPAC's case, of the 27.6m shares held by the public before the deal closed, 27.1m were redeemed. So the deal coincided with a radical drop in the float coupled with an increase in retail investor attention. Historically, those two features have predicted poor returns (albeit in a way that's expensive to take advantage of, since those stocks are expensive to borrow). SPACs just engineer that situation more frequently.
The FTX story rolls on. The company released how much it owes each of its top creditors, without identifying them; the biggest is $227m. Meanwhile, the company says that some of its regulated subsidiaries are solvent and will be sold ($, WSJ). That last detail is fascinating. Any financial institution's collapse creates the risk of contagion, but sometimes there's more contagion to counterparties outside the business than to subsidiaries within it. This is a slightly positive sign for US-regulated entities. It's still possible for them to be hit by either crypto price declines or financial ties to FTX, but the exact story with FTX, of money being moved from customer accounts to related entities that then lost that money, didn't happen in a stricter regime.
Donald Trump has been reinstated by Twitter, but says (for now) that he'll stay on Truth Social. Given that he has 4.6m followers on Truth and 87.1m on Twitter, and that many of the people he most dislikes are on Twitter rather than Truth, this may be a hard commitment to keep. It's an important one for Digital World Acquisition Corp., the company trying to take Truth Social's parent company public.
And it raises interesting questions about businesses that are dependent on a single executive. The NYT has a thorough piece on another such personality-driven brand, covering gambling company Penn National's bet on Dave Portnoy of Barstool Sports. Portnoy, like Trump, has rabid fans who drive a lot of attention to the underlying company. (In Portnoy's case, since he's a self-identified degenerate gambler and the product in question is gambling, the connection is more direct. But Twitter reliably converts outrage into ad views, and may be in much the same position.) Big personalities can draw lots of eyeballs and create plenty of short-term revenue and even market value. But the same traits that make them famous can make them infamous later, and their popularity can make it hard for a company to negotiate with them on equal footing.
Last night, Disney re-hired former CEO Bob Iger, who had stepped down last year ($, WSJ). Iger is a very energetic executive who led some major Disney changes, including its acquisitions of Marvel, Pixar, LucasFilm, and 21st Century Fox, as well as the launch of Disney+. Iger also spent a long time deciding when to retire, who his successor would be, etc. Disney shares compounded at 12.0% annualized during Iger's tenure, compared to 8.7% for the S&P; since he left, the market's up 32.8% and Disney is down 25.4%.
Media companies seem to have a disproportionate amount of office drama relative to other industries. There are stories of messy successions in other fields, but there's a reason Succession is about an entertainment conglomerate and not, say, a tractor company. The Murdochs, Redstones, Maxwells, etc. provide fodder for many seasons of high-stakes corporate drama. One reason is, of course, that the industry that reports on this kind of drama is the media industry, and it's naturally interesting to them. And owning a media outlet means having an opportunity to encourage unfavorable coverage of competing media outlets; there are many companies that earn money from targeting ads based on personal data, but only one of those companies owns The New York Times and has the attendant power to affect public perception of how that model works and who uses it.
Another possibility is that the media industry is a more deal-driven one than others, because media companies constantly exist in an uncertain state. Sometimes there's a bull market in content, sometimes in distribution, and the businesses that thrive in that dynamic are the ones able to scale up quickly, through either hiring or M&A, when the pendulum swings. Deal-heavy industries will have dense social networks where gossip spreads fast, and where sharing negative stories about a competitor can be an advantage. So the industry has high-visibility executives and more than its fair share of boardroom coups.
Shareholder activist season is upon us, and new formats for soliciting shareholders' votes will make it cheaper and easier for activists to launch proxy fights ($, WSJ). This will be especially fun to watch in the tech space. Every company that was getting valued on growth a year ago has had to rethink its spending priorities, but there's a wide range of plausible outcomes. The companies that view 2022 as an aberration will have a very different margin profile from the ones that view the previous ten years as the outlier, and there are many fading growth companies that could slam the brakes on customer acquisition and new features in order to get to cash flow positive soon. Some of those companies have already been acquired by private equity firms, and many more of them will hear from activist investors soon.
Companies in the Diff network are actively seeking talent! If you're interested in exploring growth opportunities at unique companies, please reach out. Some top current roles:
- A company bringing machine learning tools to everyone is looking for experienced ML engineers with strong product sense. (NYC)
- A well funded early stage startup founded by two SpaceX engineers is building the software stack for hardware companies. They're looking for a backend engineer who can build services that quickly process large amounts of data. (Los Angeles)
- A new service that's trolling the dating market with a better product and better monetization is looking for a full-stack founding engineer. (Los Angeles)
- A high-growth company founded by early Doordash employees is looking for a Director of Partnerships who can help design the restaurant experience of the future. (Bay Area)
- A crypto data company is looking for a frontend developer who can build powerful interfaces for working with large amounts of data. (US, remote)
If you’re 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.
We're especially interested in talking to companies interested in remote contractors with finance and operations experience. If your startup needs half of a Head of Strategy or Director of Finance, please reach out.
This can be even more important in delivery, not just for the customer but for the restaurant. Restaurants face complaints when food is late, or when it arrives cold, and they know that working with a delivery service means putting their brand in someone else's hands. But most restaurants aren't equipped to build their own logistics system. Full-stack delivery services offer comprehensive outsourcing, where the restaurant's interaction ideally starts with an order and ends when the delivery person picks it up. (Though it may also include fielding a call from an irate customer.) Other services offering the software layer can only have simple user interfaces with the important questions—where is the food and when is it getting to where it's supposed to be?—by handling complexity server-side. ↩