Slopocalypse
I compare OpenAI and Anthropic before their IPOs and rehash the shovelware debate.
Welcome to this week’s edition of The Macro Obsession.
The best round-up of current events and trends in finance, tech, and the real economy currently in your inbox!
Issue #50—Week of June 15th, 2026
Shovelware a la Claude (Redux)
Bot vs. Human
Input/Output
The AI Caste System
Hey folks!
Just a heads up, I’m out traveling. I return tomorrow, so if you sent me an email/commented/etc., you’ll hear back from me Tuesday!
My wife and I are having a blast in Northern Virginia with our family—my niblings are always so much fun. In the midst of the graduation and birthday chaos, we took a trip to DC!
I tried (and failed) to ask Secretary Wright over at the DoE for oil trading tips, made a pilgrimage to the original Washington Monument,1 and took my nephews to see one of the coolest fossil collections on the planet.
Alright, onto the newsletter!
Shovelware a la Claude (Redux)
Back in TMO #33, I wrote about one of the consequences of AI: shovelware.
For the uninitiated, this is the equivalent of slop for software development. In that story, I discussed my fear of shovelware becoming normal. We will drown in slop if this trend is left unchecked. Here’s the star of the show from that issue, showing that Claude is a major source of shovelware.
At the time, I made the argument that shovelware’s existence doesn’t validate the AI use case for coding. Nor does it mean that human coding is dead. I argue that looking at the amount of code generated is shortsighted. It tells us nothing about its use.
Claude was hitting 135,000 commits per day back in February when that data was taken—I’m sure it’s higher now—but that doesn’t tell us anything about whether they’re good or not. What if we’re just producing slop?
Well, FT has our back. Take a look at this. App reviews are down ~25% in the last two years, and the heavily used app count has fallen (we are using fewer apps in general)—all while the amount of apps released has nearly tripled.
It’s concerning that reviews are down so much. I’d expect to see a rise in reviews considering that there are just more apps. But more choices has meant less usage of multiple apps; there is a consolidation of “apps with significant usage” happening.
I argued a while ago on Seeking Alpha that Uber (UBER 0.00%↑) was trying to build the American Superapp [GIFT LINK], and this kind of behavior is exactly what I’d expect if we’re seeing progress there. A consolidation of app usage could be unrelated to the ongoing slopacalypse.2
This new data is confirmation—there is no indication that people are using more apps despite there being more apps. It’s a reverse Jevon’s Paradox (Wikipedia) of sorts.
Think TMO is neat? Send it to a buddy!
Bot vs. Human
This is a depressing stat, and now it’s official, there is more bot web traffic than human traffic. This is the Cloudflare (NET 0.00%↑) Radar, which monitors traffic across web surfaces.
Last week, it became official that the bots outnumber the humans.
This makes sense to me; there are a lot of people asking bots to go onto the web instead of themselves. It’s a double-whammy. Go to ask ChatGPT a question and you end up having it crawl several websites for you, and then feed you an answer. You never left the window, it generated web traffic.
The question is how this will skew our web data moving forward. I expect this problem to worsen over time as search is now also becoming a way for AI to crawl webpages. That means that we should expect the Internet to increasingly become a place made for bots, not humans.
I suspect that this is why large corporations put out listicles on their websites. Humans won’t trust a list of “10 Best Music Apps” put out by Spotify on their website, because of course, Spotify is always the top pick. But it doesn’t matter because those were not written for us. They were written for something more naive about human intentions.
More data is needed; I will report back when I find it.
Input/Output
Now that SpaceX (SPCX 0.00%↑) has gone live, the big AI labs are next. I find the comparison between the two, OpenAI and Anthropic, fascinating.
Here’s a few I don’t think most folks have seen. The first is revenue, although the stat I am most interested in getting once they go public is the net income.3
Anthropic has pulled ahead, at least for now, although I expect that is largely due to Meta (META 0.00%↑) and their bad habits; see “Distilled in Menlo Park” (TMO #41) for more on that.
What’s interesting is how efficient Anthropic has been, keeping their training costs way less than OpenAI, not only in the here and now but in their projections too. OpenAI is planning to hit >$100B/yr in training in the next few years, while Anthropic never plans to break half that figure.
But at the end of the day, all that matters is the ability to generate users. Since none of it is profitable, the only thing that makes sense to chase is market share.
OpenAI still handily wins here. Not only that, but they are setting user acquisition records. For the people who believe that AI is eating software, this is a scary stat.
The only thing to compete on now is IPO size. I have a feeling that the market will be kinder to Anthropic than OpenAI. I won’t be betting on either, at least not until I see their finances in official filings, but I wouldn’t bet against them.
SpaceX hit the $2T mark on its first day of trading. I wonder how long it will take these two.
More Below, But ICYMI
The AI Caste System
One thing I’ve found interesting about talking to heavy AI users is their preferences for which bot and why. I’ve asked this question several times in polls at the top of TMO and we’ve had different winners each time, with ChatGPT, Gemini, and Claude all taking first at different times.
I’m proud of that, because it means that my readers have the same feeling about these things as I do: they are commodities and we should use whichever is the best at the time.
That’s not how everyone takes it, though. It turns out, you can see a difference in users when you group them by income. I’m not sure what the implications are yet, but I definitely thought it was fascinating to see.
Claude users—congrats—you make more than ChatGPT users on average.
For the record, I’m not surprised that Meta AI users are AI’s untouchables.
I wonder how this would change if we had the Chinese models in there as well. Why are we always tracking the wrong models (TMO #18)?
See you next week.
Thanks for reading.
I wrote my history bachelor capstone on the historical deification of George Washington and the impact those efforts have had in public memory, so for me this really was a pilgrimage to come see it.
This statue, of course, is the most on-the-nose example we have, but stories about his teeth being made of wood (they were pulled from slaves and fashioned into dentures), his inability to tell lies (he was an expert at military deception and crossed lines that he was bound to by honor—and in the modern day, the Geneva Conventions—frequently in order to win, e.g. Trenton), and others are all great examples of us missing the man for the myth.
Trademark pending.
I would normally say “profit,” but I don’t expect these firms to have any profits to show.