Gonzo Startups

In this issue:

• Gonzo Startups—Is Cluely a software company or a media company? Or just an admission that when capital is abundant and can increasingly be used as a substitute for human effort, the scarcest thing for businesses is attention?
• Don't Bet on a Third World War—Prices aren't probabilities when the event being priced includes the risk of one side of the trade or another getting vaporized.
• Perplexity—Apple negotiates with one search provider by talking about buying another.
• Equity Compensation—Conditional partial reversal of older investments.
• Hiring—An offer compelling enough that step one is convincing the recipient it's not a scam.
• LLM SEO!—Information sources naturally get commercialized, whether because they run ads or because advertisers use them as a substitute for ads.

The Diff June 23rd 2025 11:34 • Click to play audio

Gonzo Startups

You could, if you wanted, create a product like Cluely that was designed for entirely straightforward purposes and then get shocked that anyone would use it for something nefarious. Every so often, this happens in other fields—well-meaning civil libertarian technologists often find out that the average user of an encrypted communications and anonymous payment protocols is using them for some pretty nefarious things, and that even if "terrorists and drug lords are getting away with X" is sometimes a pretense for broadly violating people's privacy, it's also true that terrorists and drug lords exist. You could imagine a near-possibly Cluely that's just an app-agnostic copilot that constantly provides helpful hints, less a way to cheat and more a way to play life with tutorial mode enabled at all times.

Cluely is obviously not that: their marketing is gleefully fixated on all the ways you can use their product to subvert norms, by cheating at job interviews and lying on dates, and their manifesto literally opens with "We want to cheat on everything," though by the end they walk it back a bit: "And yes, the world will call it cheating. But so was the calculator. So was spellcheck. So was Google." Which is a fun way to play semantic games—spellcheck was obviously not seen as "cheating" when it meant that workplace memoranda had fewer embarrassing typos and that the authors weren't spending much time paging through a dictionary to make sure they spelled "accommodate" correctly or whatever. But if you're taking a second-grade spelling test and you smuggle in a smartphone, you are most definitely cheating even though the same tool is innocuous in other contexts.^[1]

Cluely doesn't just produce software products, of course; they're also in the business of manufacturing a nonstop stream of headlines. Their product got the founder an internship, which he lost after uploading a YouTube video showing he'd cheated during the interview! One founder got suspended by Columbia for this kind of behavior! His video hyping a $15m fundraise from a16z mostly consists of references to a movie that came out when he was six years old! They're hiring fifty marketing interns, featuring yet another well-produced video.

At some point, you have to wonder if $15m was funding an AI wrapper business or a clever media project satirizing the occasional excesses of startupland. The entire Cluely ethos seems to be torn between a) an earnest view that reciting an AI-generated answer to a job interview question about implementing singly-linked lists is a socially-valuable activity, and that more of society's resources should be allocated to people who can do this with the appropriate level of charisma, and b) a self-aware attempt to always be about six months ahead of social norms around technology use. (There was a time when it was at least a little rude to check text messages during a social outing. Then again, there was a time when it was rude for a non-smoker not to have an ashtray or two for guests. Norms shift!)

Like many other phenomena, the celebrity CEO is older than it looks, and that celebrity is rarely a completely organic phenomenon: Rockefeller hired a PR person, Ivy Lee, who suggested humanizing stunts like handing out dimes to children as a way to make Rockefeller look more generous than rapacious.^[2] Edison was a prolific inventor, but his PR people had a tendency to take messy stories about collaborative efforts and streamline them into tales about lone genius inventors. Henry Ford, who was selling to an early version of the mass-affluent audience, was good at sculpting media narratives about his business decisions (his wages were high, but this was in response to market pressure from the nature of the work), and also enjoyed the occasional attention-grabbing stunt, whether that was the peace ship or some more dubious media endeavors.

A good argument against aggressive, not-entirely-honest marketing is that it produces a temporary burst of attention but either requires real substance or a follow-up stunt that's even more inflammatory. Arguing against this is a look at the list of the world's wealthiest people, where both Forbes and Bloomberg agree that the #1 spot belongs to one Elon Musk. Musk is a much better engineer and business operator than his critics give him credit for, and a much more deft manipulator of public and media perception than his fans want to admit. Fairly early in his career, Musk allocated 10% of his portfolio to a McLaren, which he promptly wrecked (it was uninsured). A very bad capital allocation decision, except that it meant that from then on, anyone negotiating with him knew that they were dealing with someone who had an alien utility function and an indifference to the risk of a drawdown to zero. Musk's career is obviously not entirely PR; it happens that he's chosen a model where PR can multiply the impact of other things he's good at.

So: Cluely. The business is noteworthy because it's transgressive, and the only PR strategies available are to say "it's not actually transgressive" or "it's more transgressive than a repressed square like you could possibly imagine. Deal with it." They've chosen the latter, which means that, relative to their accomplishments, Cluely has the best-marketed job postings imaginable right now and can easily attract an audience of hundreds of thousands or millions of people to whatever product announcement they dream up next.^[3] They're selecting against plenty of prospective employees and partners, too, in a way that constrains them in the long term.

But this kind of presentation is too easy to dismiss. Sometimes, the hucksters really do know what they're doing. Early in his career, then-oilman Philip Anschutz had a problem: he'd recently bought a bunch of oil fields, and they were now on fire. So, he called a famous oil well firefighter, Red Adair, and asked him to deal with the problem. Red Adair called friends in the oil industry, who told him (accurately) that Anschutz was tapped out. But Adair was famous enough that there was a movie inspired by his work that was in production at the time. So Anschutz sold Universal the rights to film his burning oil fields, and used the money to pay Adair to put the fires out. Basically every detail in this story—letting the oil catch fire, hiring someone famous enough that John Wayne was playing him in a movie, selling media rights to a disaster in order to avert the disaster—is a massive red flag. Anyone who refused to ever do business with Anschutz again on the basis of this story made a respectable decision. And yet, our protagonist ended up making a fortune (he parlayed oil money into railroad ownership, and railroads into rights-of-way for fiber optics).

It's a necessary feature of a post-scarcity economy—one where one of the biggest blockbuster products around is a drug that helps people consume less food—that attention is immensely valuable. At scale, that takes the form of converting attention into ad dollars. But for smaller companies, there's a different attention economy: in a system where every feed is ruthlessly ranked and every competition has exacting, quantifiable rubrics, there's something intrinsically exciting about building a workaround, even if it won't necessarily work that much butter. It will at least work differently, and if everyone's relentlessly optimizing along legible axes and thus implicitly optimizing against the illegible ones, even some diversity of pathologies is useful.

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

• Ex-Citadel/D.E. Shaw team building AI-native infrastructure that turns lots of insurance data—structured and unstructured—into decision-grade plumbing that helps casualty risk and insurance liabilities move is looking for a data scientist with classical and generative ML experience. (NYC, Boston)
• Deerfield-backed, Series A company building agents for healthcare administration (prior authorization, eligibility checks, patient scheduling) is looking for a senior software/AI engineer to build backend services and LLM agents. Experience building and monitoring production-quality ML and AI systems preferred. (NYC, Hybrid)
• Well-funded, fast-moving team is looking for a full-stack engineer to help build the best AI powered video editor for marketers. Tackle advanced media pipelines, LLM applications, and more. TypeScript/React expertise required. (Austin, 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

Don't Bet on a Third World War

Leaving aside the direct impact, what should an impersonal, impartial observer make of the fact that the US bombed some of Iran's nuclear weapons facilities? If futures markets are any guide, the Third World War is slightly closer, though no more closer than other geopolitical surprises implied. A fun question to ask about all of this is: what has actually happened to the world's risk tolerance?

At a surface level, this is a risk-on event, with the US launching strikes against, and revealing new capabilities against, a county that isn't quite aligned with the usual axis of enmity. But at another level: the near-term risk of some form of tit-for-tat retaliation is higher, but the long-term risk of another nuclear power is lower.

Which of these is more important? As The Diff has long argued, the optimal trade in a nuclear war is to sell, unless you're the target, in which case you should buy. So, at least in this scenario, asset prices and their volatility do not fully reflect changes in the state of the world, because the specific state that's less likely is one where those cleverly-timed put option purchases are worthless due to counterparty risk, not misjudging fundamentals.

Perplexity

Apple is considering, at least vaguely, a purchase of AI search engine Perplexity. This is always possible, but Apple has a preexisting relationship with an increasingly AI-powered search engine, and the main question in that relationship is how much each side brings to the table. So Apple's incentive is, always, to sound more enthusiastic about search-related hiring and M&A than it is, and to leak evidence of such enthusiasm widely.

Equity Compensation

Revolut's CEO may earn billions of dollars in equity compensation if he can increase the company's market value from $45bn to $150bn ($, FT). Given that he's already worth $7bn, and that money is in Revolut already, it obviously looks a bit excessive to offer him even more equity. On the other hand, once someone's worth that kind of money, it takes a lot of incremental compensation to even move the needle. One last way to look at this is that it's an effort to recut a previous deal: Revolut raised money when valuations were lower, especially for European startups, and as a result they've been diluted more than a company with a similar trajectory would be today. A massive stock grant is basically a slow-moving, tax-inefficient way to push their cap table back to what it otherwise would have been. And of course investors who own something that tripled in value don't have much to complain about if their cut of that growth is a little lower, but the growth is a little higher.

Hiring

The WSJ has more details on Mark Zuckerberg's hiring spree ($, WSJ). Every hiring approach comes with tradeoffs, and one of the ones Meta is aware of is that basically everyone in AI is constantly getting approached with interesting opportunities to jump ship. One way Meta can stand out is by having Zuckerberg reach out personally, but: "Some of the people who have received the messages were so surprised they didn’t believe it was really Zuckerberg. One person assumed it was a hoax and didn’t respond for several days." This ends up leaning in to one of Meta's core ideas: part of what set the company apart early on was its insistence that every account was associated with real names—it was a place for people to exist online, not a place to play a character online (though of course there are plenty of people whose job is, at least in part, to play a character who happens to share their real name). As content gets cheaper to manufacture, connecting it to real identity becomes more valuable, especially when the content in question is a suggestion to take a new job.

Disclosure: Long META.

LLM SEO!

There are now companies that help their customers understand and control what chatbots are saying about them. Some of this is obviously negative for the quality of online content, since a hacky way to do this is to just astroturf Reddit (at least unless Reddit starts using Worldcoin's biometrics to verify users' identities). But it also encourages more transparency: the LLM product research experience is pretty good, but it's unclear how many products don't show up simply because there wasn't an incentive for the product's creator to post detailed information about it online. If that information could contrbute to an ultra-long-tail search that leads to a purchase, there's an ROI on being transparent.