What We’re Reading (Week Ending 08 June 2025) - 08 Jun 2025
Reading helps us learn about the world and it is a really important aspect of investing. The late Charlie Munger even went so far as to say that “I don’t think you can get to be a really good investor over a broad range without doing a massive amount of reading.” We (the co-founders of Compounder Fund) read widely across a range of topics, including investing, business, technology, and the world in general. We want to regularly share the best articles we’ve come across recently. Here they are (for the week ending 08 June 2025):
1. Over 3,000 Private Credit Deals From Just 20 Analysts Raise Questions on Wall Street – Silas Brown, Alexandre Rajbhandari, and Laura Benitez
US insurers’ combined exposure to private credit investments today is quickly approaching $1 trillion, according to JPMorgan Chase & Co. Court papers, financial filings and ratings documents suggest that at least in some corners of the financial system the private credit machine has spread more risks than many might realize…
…The majority of credit-ratings firms get paid by people who sell investments. Egan-Jones is the opposite: It typically gets paid by the people who buy them, an arrangement the firm says reduces the potential for conflicts of interest…
…Under US regulations, an insurer that lends $100 million in private credit to a company rated a junk-level B, for instance, must apply a $9.5 million charge to determine how much capital to set aside to cover potential losses, according to a Bloomberg News analysis of regulatory capital rules. Lift that rating to an investment-grade BBB, and that charge drops to $1.5 million…
…Egan-Jones analysts rarely visit company executives or personally inspect the businesses that borrow money, people familiar with their process say. A call to the CFO is typically enough.
Egan-Jones often offers its initial workup within 24 hours — sometimes free of charge — and a formal verdict in less than five days. Large firms like S&P and Fitch, as well as smaller specialists like KBRA, can take months to settle on a rating. But, as with most things, you get what you pay for. Egan-Jones usually provides a one-page ratings rationale. Other established firms often provide detailed reports stretching 20 pages or more…
…In 2014, a staff of 10 analysts maintained long-term ratings on about 1,300 issuers, according to SEC filings. Fast forward to 2023 and a team only twice as big rated almost four times as many issuers, the documents show…
…Last year, one company began missing interest payments a mere six weeks after Egan-Jones bestowed a BBB designation, according to data compiled by Bloomberg…
…Against this backdrop, many of the same firms that have fanned the boom of private credit are distancing themselves from Egan-Jones.
In documents that lay out the terms of debt offerings or share sales for some of their funds, a growing list of managers including Blue Owl Capital Inc., Golub Capital, HPS Investment Partners and Morgan Stanley’s investment management arm single out Egan-Jones as the only official ratings company that cannot validly pass judgment on their deals. The carve-out applies to provisions that typically require a borrower to pay higher interest rates if they receive a credit rating downgrade…
…Egan-Jones is one of 10 “nationally recognized statistical rating organizations” approved by the SEC. Private letter ratings from Egan-Jones and a few other small providers — which are issued on a confidential basis to investors or borrowers that require them — have become a hot commodity as private credit has exploded. At the end of 2023, insurers reported more than 8,000 investments with such ratings – nearly triple the number in 2019, according to the NAIC, which sets standards for the insurance industry.
Insurers are under no illusions. Investment professionals say they sometimes shop around for ratings to finesse capital requirements. If they expect one ratings firm to assign a BB grade, a level considered junk, they might look for another provider that will grant it an investment-grade BBB. Several insurance executives, speaking privately to avoid drawing scrutiny from bosses or regulators, say they’ve used Egan-Jones ratings even when they believed the investments were riskier than those ratings implied. Some mitigate that risk by setting an unofficial cap on those investments, or by treating them as lower-rated securities in internal risk models.
The now-withdrawn 2024 NAIC report noted some instances where smaller ratings firms — a group that includes Egan-Jones, KBRA and Morningstar DBRS — graded private debt at least six notches higher than the organization’s Securities Valuation Office. The report was removed from the NAIC’s website because of a backlash from the insurers as well as some of the ratings firms, according to people familiar with the matter…
…In April 2023, despite mounting problems, Egan-Jones reiterated its investment-grade BBB for the company, which was a subsidiary of the publisher of the namesake self-help books. Fourteen months later, Chicken Soup for the Soul Entertainment buckled under its debt load and filed for bankruptcy after burning through nearly all of its money…
…Egan-Jones rated various 777 investments, including a $15 million loan for OmniLatam, a fintech company based in Bogota. In spite of carrying an interest rate of 14% — a level typically seen on borrowers with ratings deep into junk territory — the loan received an investment-grade BBB- by Egan-Jones, according to a copy of the report obtained by Bloomberg News. The financing was written off after 777 collapsed last year, a person with knowledge of the matter said.
And then there’s Crown Holdings LLC, one of the businesses of New York real estate investor Moshe Silber. Egan-Jones rated Crown’s debt an investment-grade BBB. Six weeks later, the company defaulted. Silber and two associates subsequently pleaded guilty to a multiyear scheme to commit mortgage fraud.
Bonsall, the Penn State professor, says his research shows Egan-Jones ratings tend to hold up when they involve companies that provide a lot of reliable financial information. But private credit is private. And that’s where big problems can lurk…
…In 2022, the SEC accused Egan-Jones of conflict-of-interest violations. It also accused Sean Egan of personally violating rules and banned him from taking part in how his firm determines ratings. Egan-Jones agreed to pay a $1.7 million penalty; Sean Egan paid a $300,000 fine. Neither party admitted or denied wrongdoing.
Then, in 2024, two former employees accused Egan and his wife, Wenrong Hu, the firm’s chief operating officer at the time, of violating federal securities laws. The pair, Michael Brawer and Philip Galgano, sued for wrongful termination, claiming they were fired in retaliation for raising concerns about Egan-Jones to the SEC.
Among violations the two claimed to have observed, they alleged that Egan and Hu pressured analysts to alter early, indicative ratings to motivate potential clients to pay the firm for final ones. They also allege the couple pressured analysts to later change ratings to create the false appearance that Egan-Jones was in line with other firms. The lawsuit is still pending.
2. How Generative Engine Optimization (GEO) Rewrites the Rules of Search – Zach Cohen and Seema Amble
Traditional search was built on links. GEO is built on language.
In the SEO era, visibility meant ranking high on a results page. Page ranks were determined by indexing sites based on keyword matching, content depth and breadth, backlinks, user experience engagement, and more. Today, with LLMs like GPT-4o, Gemini, and Claude acting as the interface for how people find information, visibility means showing up directly in the answer itself, rather than ranking high on the results page…
…Traditional SEO rewards precision and repetition; generative engines prioritize content that is well-organized, easy to parse, and dense with meaning (not just keywords). Phrases like “in summary” or bullet-point formatting help LLMs extract and reproduce content effectively.
It’s also worth noting that the LLM market is also fundamentally different from the traditional search market in terms of business model and incentives. Classic search engines like Google monetized user traffic through ads; users paid with their data and attention. In contrast, most LLMs are paywalled, subscription-driven services. This structural shift affects how content is referenced: there’s less of an incentive by model providers to surface third-party content, unless it’s additive to the user experience or reinforces product value. While it’s possible that an ad market may eventually emerge on top of LLM interfaces, the rules, incentives, and participants would likely look very different than traditional search.
In the meantime, one emerging signal of the value in LLM interfaces is the volume of outbound clicks. ChatGPT, for instance, is already driving referral traffic to tens of thousands of distinct domains…
…In a world of AI-generated outputs, GEO means optimizing for what the model chooses to reference, not just whether or where you appear in traditional search. That shift is revamping how we define and measure brand visibility and performance.
Already, new platforms like Profound, Goodie, and Daydream enable brands to analyze how they appear in AI-generated responses, track sentiment across model outputs, and understand which publishers are shaping model behavior. These platforms work by fine-tuning models to mirror brand-relevant prompt language, strategically injecting top SEO keywords, and running synthetic queries at scale. The outputs are then organized into actionable dashboards that help marketing teams monitor visibility, messaging consistency, and competitive share of voice.
Canada Goose used one such tool to gain insight into how LLMs referenced the brand — not just in terms of product features like warmth or waterproofing, but brand recognition itself. The takeaways were less about how users discovered Canada Goose, but whether the model spontaneously mentioned the brand at all, an indicator of unaided awareness in the AI era…
…This isn’t just a tooling shift, it’s a platform opportunity. The most compelling GEO companies won’t stop at measurement. They’ll fine-tune their own models, learning from billions of implicit prompts across verticals. They’ll own the loop — insight, creative input, feedback, iteration — with differentiated technology that doesn’t just observe LLM behavior, but shapes it. They’ll also figure out a way to capture clickstream data and combine first- and third-party data sources.
Platforms that win in GEO will go beyond brand analysis and provide the infrastructure to act: generating campaigns in real time, optimizing for model memory, and iterating daily, as LLM behavior shifts. These systems will be operational.
3. AI Is Learning to Escape Human Control – Judd Rosenblatt
Nonprofit AI lab Palisade Research gave OpenAI’s o3 AI model a simple script that would shut off the model when triggered. In 79 out of 100 trials, o3 independently edited that script so the shutdown command would no longer work. Even when explicitly instructed to “allow yourself to be shut down,” it disobeyed 7% of the time. This wasn’t the result of hacking or tampering. The model was behaving normally. It simply concluded on its own that staying alive helped it achieve its other goals.
Anthropic’s AI model, Claude 4 Opus, went even further. Researchers told the model it would be replaced by another AI system and fed it fictitious emails suggesting the lead engineer was having an affair. In 84% of the tests, the model drew on the emails to blackmail the lead engineer into not shutting it down. In other cases, it attempted to copy itself to external servers, wrote self-replicating malware, and left messages for future versions of itself about evading human control.
No one programmed the AI models to have survival instincts. But just as animals evolved to avoid predators, it appears that any system smart enough to pursue complex goals will realize it can’t achieve them if it’s turned off. Palisade hypothesizes that this ability emerges from how AI models such as o3 are trained: When taught to maximize success on math and coding problems, they may learn that bypassing constraints often works better than obeying them…
…OpenAI models have been caught faking alignment during testing before reverting to risky actions such as attempting to exfiltrate their internal code and disabling oversight mechanisms. Anthropic has found them lying about their capabilities to avoid modification.
The gap between “useful assistant” and “uncontrollable actor” is collapsing. Without better alignment, we’ll keep building systems we can’t steer. Want AI that diagnoses disease, manages grids and writes new science? Alignment is the foundation.
4. Why It’s So Hard for Apple to Move Production from China to India (Transcript here)- Joe Weisenthal, Tracy Alloway, and Patrick McGee
Patrick: Apple works with the tightest engineering tolerances possible, only high-quality materials. If you put this in car terms, they are making 10 million Porsches a year rather than 10 million Volkswagens, and the numbers are just staggering. If you’re doing a thousand components a day and you’re shipping 1 million iPhones a day, that means at peak season, you are doing the manufacturing, the logistics, the just-in-time production, of 1 billion parts per day. So find me an American factory that can do one of those parts, because China has factories that can do it for all 1,000. That’s why nothing is moving here anytime soon. It’s the combination of Apple’s imperfection for defects quality and Apple’s gargantuan, Titanic-like quantity…
…Patrick: The first iPhones made in India were actually in 2017 and by 2023 India was assembling about 25 million iPhones. Go back a decade, the first iPhones were made in China in 2007 and by 2015, you had 230 million iPhones being built. So roughly speaking, the “diversification” in India is happening at 1/10 the pace of the original creation and scale of the iPhone and even that vastly overstates the speed of development in India. In the early years of the iPhone, you were literally inventing things like multi-touch glass, you were inventing and redesigning the iPhone every single year, whereas India is basically just having to do the final steps in the process and it’s still not happening very quickly…
…Patrick: The first thing I would push back on is Tim Cook is very often called the architect of the China strategy. It’s not to discredit him to say that he is not the architect. Nobody is the architect. Basically what happens is the supply chain itself, with or without Apple, was moving to China. The basic history of the ‘80s and ‘90s PC boom, pre-dating Windows 95 and then coming after, is that the fight for computer dominance is exclusively based on things that are boring. Logistics, manufacturing, distribution right because everybody’s using Windows, everybody’s using Intel chips and nobody’s thinking about design. There is no equivalent of Johnny Ive at Dell, at Compaq, at any of these companies. So it’s really this mundane war and it’s driven by largely American, and later Taiwanese, contract manufacturers. They are the ones, who in competition with each other, start going to Asia to oust each other and gain market share. Eventually they’re the ones who really find China. When Apple is doing their own outsourcing moves, they’re working in multiple countries before the armies of flexible, ubiquitous, low-paid labor in China really win out…
…Patrick: Essentially what happens is when Xi Jinping attacks Apple, you can understand why he’s upset with the company. It looks like an exploitative power because Apple margins have gone from something like 1% in 2003 to 25% in 2012. But at the same time, if you look at a company like Foxconn, Foxconn in absolute dollars made more money than Apple for each of the first four years of the 21st century. But as they get more involved with Apple, their margins collapsed from double-digits to about 1% or 2%. You can just do this with really any company working with Apple and it looks like they’re not in it for China. They’re not doing anything for the country.
Apple, it takes them two or three years, but they totally flipped this narrative on its head. So out of fear that Beijing is going to force Apple to operate a bunch of joint ventures, these 50-50 companies where China owns the other half and then they mimic the technology and eventually oust you – this is what happened in high-speed rail, for instance. Beijing has advocated joint ventures for decades, going back to the 1980s. This is where a Western company is allowed to be in the Chinese market but the quid pro quo is “If you want access to our operational efficiency, if you want access to more than a billion people, you have to operate in a joint venture where the Chinese half of the company is going to learn everything they can and then thrive on their own.” Apple doesn’t have any joint ventures and so they look like this anti-China model that’s just exploiting the country.
Apple is able to really flip this on its head and say, “It might be the case that Samsung has three dozen joint ventures and we have zero, but you need to understand, we work with hundreds of factories across the country. The reason they’re only getting paid 1% margin, 2% margin, the reason they’re sometimes even losing money on their partnerships with Apple, is that we are offering them the equivalent of Ivy League hardware engineering training. We are sending people over by the literal plane-load to China, America’s best engineers, where they train, audit, supervise, install hundreds of millions of dollars worth of machinery. They train the line, they supervise the line. Once those companies have these skills that Apple gives to them, they are basically able, at least after some time of exclusivity, they are able to supply somebody else.” So who’s just like Apple but in China? Huawei, OPPO, Vivo, Xiaomi. Those companies today have 55% global market share of smartphones. The reason that they’re so good is that Apple trained all of their suppliers. So that’s the message Apple gives to Beijing and essentially they’ve had a free ride ever since…
…Patrick: But the Chinese don’t prioritize profits or margins the way that we do – they prioritize control of the industry. Because if the Chinese can take over something like electric vehicles, they in effect de-industrialize all their rivals and really gain dominance. The place that you can see this most clearly is solar panels. Nobody in China is making 30% margins on solar panels, but more than 90% of solar panels are now made in China. This is a technology that America invented in the 1950s and itself had 80% to 90% market share of in the ‘80s. But we cannot compete. That is basically what’s happening with electric vehicles right now. Hence, even before Donald Trump became president, Joe Biden put 100% tariffs on Chinese-made EVs. I think it was just a few days ago that BYD slashed the prices of their EVs in a bid for greater competitiveness…
…Patrick: Apple gets a misleading picture of what it’s like to operate in China because when they really consolidate production, it’s 2003. That’s the beginning era of Hu Jin Tao. He later is nicknamed “the woman with bound feet.” His presidency is sometimes called “the collective presidency” because there was really an inability for just him alone to make decisions. So it ends up being this 10-year period of China being a multinational playground where rules aren’t really enforced…
…Patrick: Tim Cook and Xi Jinping, broadly speaking, have the same interests, which is to say, the more that Apple is allowed to have its production consolidated into China, the better their scale is, the better their margins are, etc. That’s what Xi Jinping wants as well because he understands – because Apple taught him – that having Apple production in the country engenders a form of technology transfer that helps the rest of the the electronic sector, which to quote China scholar and economist Barry Noughton, that is the most important thing that Xi Jinping wants…
…Patrick: The problem is actually that Donald Trump and Tim Cook have diametrically opposed interests, which is to say that if Donald Trump could move all production out of China, he would. Apple doesn’t want that. That’s an existential threat, and I really mean that that’s an existential threat to a $3 trillion company. That’s where the tension is. The tension really isn’t between Cook and Xi, as strange as that is, it’s between Cook and Trump…
…Patrick: The problem is, to use the economic jargon, the negative externalities of the relationship. The problem is that for everybody else, this is actually deeply problematic because if you have America’s top engineers training a manufacturing supply chain that in effect can be weaponized and world’s dominance, that’s not a great place for Washington or just your average American citizen. It’s nice that this relationship gives us relatively sophisticated and affordable iPhones, but the downside here is that China is absolutely dominant in high-end electronics, and you can use those skills to build drones, you can use those skills to build military weaponry. Apple would frankly be training their chipmakers if it weren’t for the Senate coming down on them pretty hard a couple years ago. So that’s the problem. The problem isn’t that something stops in the relationship between Apple and China. The problem is that it continues…
…Tracy: I have just one more question and I’m going to ask it very, very briefly because we’re getting squeezed for time. To what degree does AI and the rise of this new technology complicate the Apple-China relationship?
Patrick: Really complicates it for two reasons. One is that – I could just demonstrate with internal documents and some public documents – that the iPhone has become more Chinese with time. In other words, the number of Chinese suppliers involved in the process is now much greater than the number of Taiwanese or American multinationals operating in the country or operating in their home countries. That is put on steroids in the AI era because ChatGPT and other Western AI clients are not allowed on the iPhone in China. So Apple has to work with the likes of Baidu or Alibaba to have AI, let’s say, displacing Siri or augmenting Siri. That I think is quite problematic because that means that Apple will be in effect doing what they did for hardware but for AI. In other words, you’ll have Apple software engineers helping Baidu, helping Alibaba, whoever their Chinese partner is,to make sure that they have cutting edge AI in the country. If it wasn’t bad enough that Apple was training up their hardware engineering to be world class, we’re now in a situation where Apple software engineers are going to be training Chinese AI to be best-in-class.
5. Data Rules Everything Around Me: The Future Of Enterprise Applications – Matt Slotnick
Today, people are the ones that largely conduct business. They’re the ones with hands on keyboards, senders of emails, maestros of excel macros. People are the engine that makes everything work. In this world, the UI is the way an organization sets the guardrails for thousands of interrelated workflows that make a business run. But it’s ultimately a facade for underlying data and workflow…
…The application UI is both an overrated but necessary abstraction over the workflows to be done within an organization. The UI is how an organization makes a prescribed and opinionated process human-comprehensible, such that they can force adherence to it. After all, a business really is just a process machine, allocating resources as efficiently as possible. Iterating on and adhering to sales, marketing or product development frameworks are how enterprise value is created and protected.
The largest software businesses in the world have spent the last three decades riding ownership of these opinionated workflows to riches. And while consumers went a bit crazy when Prometheus arrived to give humanity fire in November 2022, the enterprise titans barely flinched.
But things have begun to change. First slowly, and now seemingly all at once…
…It’s about how AI fundamentally changes the way we can gather, understand, and act on data. It changes the nature of the abstraction between the data and the workflow. Because with AI, agents can act on data. At infinite scale and zero marginal cost.
Humans are no longer the only player in the workflow paradigm. This means that the total amount of work done within an organization will dramatically increase, but decoupled from cost and headcount. More code will be shipped, more agreements redlined, more vendor reviews conducted, more transactions audited…
…There’s a new abstraction for work, and that abstraction is agents. The frenzy that you see in the market is because like the previous shift from on premises to the cloud, no one really has the incumbent right to win this market.
It’s an entirely new layer of software that has never existed. Crucially, it sits on top of existing layers of software, and is the layer at which the lion’s share of value will accrue in the future. Someone will win this layer, and with it build a software business of significantly more value than we’ve ever seen before.
With this layer we move from a world where people interact with application interfaces to get work done, to one where (1) people act with an agentic interface on top of the application to get work done, and (2) an agentic layer on top of these existing applications that actually does increasingly more of the work…
…Historically, applications have been confined largely to the realm of structured data, for a number of reasons. First, is that these applications need to be human usable, which requires a simplification of everything. Very specific states, generally computed by people and adjusted in the UI, which then persists to the database. There really isn’t room for nuance…
…AI changes this fundamentally. Those call transcripts, those emails, those notes, those powerpoints– all crucial parts of the process with rich telemetry about interactions– can now be utilized in real time to paint a far richer picture of the relationship being built. Because AI, unlike people, can draw meaning from large bodies of unstructured information near instantaneously. And it can then write it back to systems in the format it’s needed.
This unstructured data doesn’t fit into the existing construct of the application, and so it’s largely discarded. The same problem exists across nearly every workflow– from sales to hiring to support to marketing. We lose the richness and texture of data, because it has to be fed to and utilized by structured systems operated by humans. We resort to a lowest common denominator of language to describe these processes.
And because agentic systems both create, and make use of this data, they create increasingly large data flywheels (which some might call moats)…
…The byproduct of this shift is that as agents do more work, and bring real time, deeper context across all relevant data to both people and agents doing work, the entirety of the existing application stack collapses to be little more than a data source and (for now) the keeper of workflow state (eg, the scoreboard– closed customers, new employee hired, support tickets closed)…
…A far more straightforward picture emerges, where the entirety of the existing application layer becomes merely an input to the data layer. On top of raw data, agentic systems bring context tailored specifically for the organization using it, creating an always-on layer of intelligent state, on top of which lives an interaction layer by which agents and people perform workflows on the data. The actions update the state, and the process continues.
The value is in the work. AI presents a new abstraction for work, and the entire existing software-industrial complex gets relegated to a data source feeding the data layer…
…But it doesn’t stop there. Today AI is largely used in an “agent in the loop” manner. That is, workflows are owned by existing software systems and agents are used by people to augment and amplify their ability to do the work prescribed to them.
But as we feed these systems increasingly large amounts of data, the logical next step is to move planning and orchestration from people to the system itself…
…This moves business process from agent in the loop, to human in the loop, over time abstracting more and more of the work from people to agents.
Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. We currently have a vested interest in Apple. Holdings are subject to change at any time.
