|
In this issue: - Journeymen, Hobbyists, Superstars—If there's a job people will do for money, but also for fun and social status, then a more efficient labor market will tend to price out professionals and replace them with hobbyists. In media, that leads to some surprising feedback loops. It lowers the average quality of the output, ramps up the variance, and gives everyone more of what they want (which is not necessarily a good thing).
- Moving Targets—The cycle from concept to fast food dish keeps accelerating.
- AI Insurance—It's hard to insure if you don't know the risks, and even harder to insure if your counterparty can use that as permission to take more risks.
- AI Debt—Yields for AI-related bonds are up. Is it a market signal, or just market structure?
- Favors for Friends—The publicly-traded strip club chain continues to act less like a public company and a lot like a strip club.
- Alpha—Utilities as the AI volatility proxy.
Journeymen, Hobbyists, Superstars
Fifty years ago, the default media environment was that most cities had a small oligopoly of TV stations, a broader set of radio options, and a newspaper monopoly. All of those needed content to sell ads against, so there was an economic need for someone to sit in on city council meetings and public hearings, review local restaurants and shows, write up sports games, etc. That's shifted, for what's probably a combination of technological reasons (cable TV and especially the Internet) and changes in consumer taste (more interest in national and less interest in local news).
That basically amounted to a jobs program for journalists who were probably not going to be among the best in the world, but who were nevertheless pretty good at their job. It helped that it was hard to come up with a direct connection between a writer's output and their economic impact; you infer that people are reading the news when they renew their subscription, but you really have no idea whether they're reading every issue cover-to-cover, or just following sports, just looking at classified ads, stock quotes (depending on the locality), just skimming the obituaries, etc. On the other hand, readers would offer negative feedback on specific articles. So it made some sense for newspapers to hire for quality, but to be unable to figure out exactly who was contributing how much. A great recipe for a reasonably well-paid job—maybe partially paid in social status and access to interesting people—but a reasonable middle-class job nonetheless.
You don't hear too many stories about people making a middle-class income writing local news lately. The most popular writers make much more than they used to—William F. Buckley was the most popular syndicated columnist in America, but his ritzy lifestyle was funded partly through his own and especially through his wife's inheritance, while there are multiple seven-figure Substackers out there (and at least one nine-figure Substacker), and The Atlantic was apparently offering $300k pay packages to poach new writers. The journalism business is still a decent business, for a tiny number of people.
But that's partly an artifact of online distribution creating a more complete market. Traditional publications are, if you squint, a marketplace business: there are lots of people who want to read, watch, and listen. There's a smaller cohort of people who can produce something worth reading, watching, or hearing. It's a messy marketplace with slow feedback loops. But lop a few orders of magnitude off the cost of distribution, and things get more interesting: you can figure out what people like to read in the aggregate, and make more of it; you can figure out what specific audience members are looking for, and if there are enough of them, you can create it.
And the "you" in question is suddenly a broader category than people working in media. Anyone can show up and start writing, or take a video that turns out to be newsworthy, or start uploading tracks to Soundcloud, or record a podcast. And, as it turns out, if the default increment of working in a field is no longer the typical workweek of a full-time job, lots of people will do it for free. Or, to be more specific, they'll do it part-time if they're sufficiently well-paid in status, fun, and impact. When the minimum increment of delivered content is no longer a 30-minute show, an album, or a full newspaper, but instead a track, a clip, or a tweet, it's much easier to reassemble these into a user-specific feed. Posting on social media means creating an instant ad hoc publication for whatever micro-demographic is interested—check out the headline in the Everybody Who's Friends With Someone who Went to that Wild Party on Saturday Picayune or the Everyone Who's Excited That Your Baby Learned to Walk Channel.
This leads to a much more efficient market, but that's efficiency in the narrow sense that it's easier for people to get what they want, not that it's easier for them to get what they should want. Fully decentralized journalism has a disadvantage at producing lengthy investigative works; if a story might take months of full-time effort, a part-timer or a distributed team probably won't do it. Or, if they do, it will be very specific kinds of online journalism—various online subcultures will sometimes produce investigations dozens of pages long about fabulist bloggers, plagiarist fanfiction writers, speedrun cheaters, and sex pests. They produce fewer deeply-investigated stories of the "this factory is emitting ten times the legal limit of..." variety. (Though, when the company in question is publicly traded, Hunterbrook might look into it. The complete market taketh away, but it giveth, too.)
It also means that there's less disinterested journalism. In the old media era, if a war broke out somewhere there was a decent chance that there were already some journalists from big publications nearby, and more would show up to provide local coverage. Now, there are a lot more photos and videos, and depending on the conflict you can get real-time coverage if you read the right Telegram channels. But that's all coverage from people who are actively involved in the conflict; they're the ones getting bombed and conscripted, and they have pretty strong opinions. Obviously you can mentally filter what they say based on that, but you can't easily infer what they omit. Obviously outsiders can have biases, too, and may inadvertently or deliberately trade comprehensiveness for access. But they're less likely to have inherited generations-old grievances.
That wider surface area and higher granularity has another effect, though: it means that identifying talent is much less of a dice-roll. You don't have to go all-in on a media bet; you can do it part-time, and then scale up the time commitment if it works. So more people get a shot at fame, and, if they're appealing, they can get in front of a larger audience much faster without needing the approval of specific gatekeepers. The odds of a one-in-a-million talent getting discovered are a lot higher if they can be discovered directly by fans, rather than by an agent, a studio executive, or an editor.
Generative AI has a surprisingly small impact on this; the more creators go direct, the more they're getting paid for the parasocial relationship rather than the content itself. The market for custom content written in someone's voice or performed in their style is different from the market for the real thing. Genuine news is always outside of the training data; there isn't much demand for hypothetical Trump quotes relative to real ones—and when there is, the actual limiting factor is coming up with an original concept for a funny place to deploy Trumpisms. Generative AI does have the effect of separating the content from the medium; if you like a podcaster's ideas but don't like the sound of their voice, you can read a transcript (or you can feed that transcript to an audio model trained on a voice you do like).
At least in this case, the real impact is from another kind of AI: better feeds are more efficient at creating fan clubs, and it's the fan club that drives incremental content creation in a more fragmented media environment. We have a lot more media than ever before, and as with plenty of other transitions to mass production, the average quality is a lot lower than it was before. But that's just the flipside of lower costs: you wouldn't bother to spend a lot of money on production (and distribution) costs and then skimp on the quality of the content, but when those costs get crushed, the quality bar something has to clear is lower. But we haven't produced much more free time to consume it all, so the new well-targeted garbage crowds out higher-quality offerings. It's tempting to try to blame this on someone, but conveniently you can blame basically everyone. The market got more efficient at delivering both the media consumption people want and the media production they'll charge the least for. What we wanted made some jobs disappear, and replaced them with hobbies, and it made some kinds of information scarcer, though entertainment has never been more abundant. Anyone can push back against this: you just have to vote with your eyeballs and thumbs.
|
As an occasional and in this case timely reminder, The Diff has an associated AngelList syndicate, open to accredited investors. It’s free to sign up, and participation happens deal by deal.
|
|
|
|
Diff JobsCompanies in the Diff network are actively looking for talent. See a sampling of current open roles below: - 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 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)
- A growing pod at a multi-manager platform is looking for new quantitative researchers, no prior finance experience necessary, exceptional coding and stats skills required. 250k+ (NYC)
- Series-A defense tech company that’s redefining logistics superiority with AI is looking for a MLE to build and deploy models that eliminate weeks of Excel work for the Special Forces. If you want to turn complex logistics systems into parametric models, fit them using Bayesian inference, and optimize logistics decision-making with gradient descent, this is for you. Python, PyTorch/TensorFlow, MLOps (Kubernetes, MLflow), and cloud infrastructure experience preferred. (NYC, Boston, SLC)
- 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)
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
Moving Targets
The WSJ has a nice profile of Taco Bell's l Chief Food Innovation Officer, Liz Matthews ($, WSJ), and it's an interesting look at the just-in-time food trend manufacturing thesis ($, Diff). It's much easier for food trends to percolate through social media and then turn into something people search for on delivery apps, but it's also much easier for novelty products to get popular online—in fact, there's an incentive for chains to release stunt products that people will buy just so they can tell TikTok or Reels that they actually ordered it. They can essentially shift some of their marketing spend to R&D instead: they're still earning attention, and occasionally a weird novelty food item will turn out to be a big hit (as the piece notes, this has been happening for a long time: the company's origin story starts with a hamburger stand owned by one Glen Bell, who noticed that tacos were getting trendy and decided to get in on it).
AI Insurance
Insurers are starting to write policies that specifically exclude AI-related risks ($, FT). There are two stories here:
- To the extent that a company is considering an AI transformation plan, worries about the downside, and can use insurance to cover that downside, they're subjecting insurers to immense adverse selection. Insurers want to drive business growth when they allow companies to implicitly sample from the average outcome instead of risking low-probability but high-risk ones; they really don't want to be in the business of flipping something from negative- to positive-expected-value, because the gap between the two comes straight out of their P&L.
- We're also at a point where AI deployment is well ahead of the legal precedents around it. We just don't know how liability will be apportioned, and how big it will be.
One of the social functions of insurance is to tell you what risks are controllable and which ones aren't; if you can buy insurance against it (without committing fraud) then whatever it is has a substantial luck component. And another function of the insurance industry is to make things a matter of luck: if you want to insure a factory against fires and industrial accidents, the insurance company is going to want to know exactly what you do to mitigate those risks so they can sell you a policy that only applies to bad outcomes. There are literally safety technologies named for this, like the underwriter's knot and Hartford Loop. And it will be very interesting to see what standards insurers use to decide what a safe policy to write is.
AI Debt
From the lender's perspective, credit markets have the same payoff structure as insurance: you get a steady stream of income, in exchange for the risk that some unexpected bad thing will happen and you'll be on the hook for the associated losses. So it's a bit symmetric that credit markets are pulling back from AI bets at the same time that insurers are ($, WSJ). But there's also a simpler supply-and-demand issue: bond investors will often have a mandate to limit the amount they lend to any one borrower, so the more the biggest borrowers need, the more of a premium they'll pay. Oracle and CoreWeave are essentially paying lenders for three separate services: first, the actual risk that their cash flows won't be enough to pay off the debt; second, that the ride will be bumpier as more of each lender's portfolio is concentrated into a smaller set of names and an even smaller set of drivers; and third, the reputational and career risk that ensues when the newest and biggest position is the one going bad.
Favors for Friends
A company that wants to return capital to shareholders has a few options: it can pay a regular dividend or a special dividend, it can run a continuous buyback, or it can do a tender offer—a one-off buyback at a premium to the share price, which typically has some kind of auction/pro-rata mechanism to figure out how much of whose shares get repurchased. You might even make a tender offer to a specific shareholder: if the company plans to buy back stock, and one big holder wants to sell, it can be easier to just do a one-time deal. For example, on Friday a reinsurance company, Axis, agreed to buy back about 3% of its shares at $99/share, about 2% below the market price when the deal was announced.
Or, if you're RCI Hospitality (the publicly-traded strip-club rollup), you can buy a bunch of stock from one specific investor at a 60% premium instead.
Tender offers to a single shareholder at a premium used to be a pretty common corporate finance tool, particularly as a way to shoo activists away. The generous theory is that it's a way for management to get rid of a distraction, but the pretty obvious read is that management is taking money from one set of shareholders and giving it to the shareholder who wants them fired. It was unpopular enough that it's had a special 50% tax since 1987. This doesn't fit the pattern well, because the shareholder, ADW Capital, is not an activist owner, though they've been activists in other situations. ADW is, however, a big holder of a stock that isn't especially liquid, and that has been accused of evading taxes and bribing auditors. So they might have told management that they planned to go activist and taken the tender as a counteroffer.
We're at a strange point in the market where stories that should be negative for companies end up making their shares pop instead. Some very dilutive offerings get structured in a way that lets the company highlight that it's selling shares above market and hope that retail traders don't do the math on what the warrants that came with those shares are worth. This is the inverse of that. On the one hand, it's a company expressing confidence in their business by buying back 9.5% of their outstanding shares. On the other hand, the company transferred $11m of shareholders' money to a single hedge fund, which makes those shareholders that much poorer.
Alpha
The FT has a look at the stocks most shorted by hedge funds ($, FT). AI has been a consensus long trade for a while, but if you're getting paid for absolute returns then it's hard to justify your compensation if your return stream can be replicated by just buying shares of the biggest companies on earth. One way they're adjusting is by shorting more of the peripheral bets—short interest in utilities is close to an all-time high. Utilities in general are a tricky way to play AI, because buying them is a way to go long usage but to go short efficiency improvements in hardware and software. The industry's stability meant that valuations ratcheted up fast when expected growth ticked up slightly, but that also means that they're the industry that, on a risk-adjusted basis, is most levered to uncertainty about AI. If you think demand will collapse or you think there might be some radical unlock that completely changes the industry's unit economics, you can express either bet with the same short position.
|