What We’re Reading (Week Ending 15 March 2026) - 15 Mar 2026
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 15 March 2026):
1. The subtle art of not selling stocks – Chin Hui Leong
My co-founder, David Kuo, has an investing rule that some of you may find peculiar: He never sells any stock he buys…
…Before you dismiss this idea as reckless, consider what this commitment actually demands.
If you know you will never sell a stock, every purchase becomes a permanent decision. You can’t afford to be casual. You can’t buy on a whim and figure it out later…
…David treats his stock purchases the same way. By removing the option to sell, he raises the bar for every stock that enters his portfolio. The result is a collection of businesses he knows deeply and trusts completely…
..Most selling decisions are driven by emotion, not analysis. When a stock drops, fear kicks in…
…Daniel Kahneman, the Nobel laureate and father of behavioural finance, would recognise this pattern. In his parlance, your reflexive brain (called System 1), built for snap decisions and danger avoidance, often overwhelms your analytical (System 2) brain you before you have a chance to think things through…
…Back in January 2007, I bought shares of Netflix at a split-adjusted US$0.33 per share. Over the past two decades or so, the stock has soared, crashed and soared again.
Along the way, I sold half my position. At the time, it felt like the prudent thing to do. Lock in the gains; reduce risk; be sensible.
But here’s what “sensible” cost me: I estimate that the shares I sold would have gained over 14,000 percent had I held on. That’s the equivalent of holding 140 stocks that went to zero.
And the chances of finding another Netflix are slim. My remaining shares are up over 300 times my original investment. The half I kept is doing the heavy lifting: the half I sold become my most expensive lesson.
For David, his eyes are on the dividend stream his shares produce, not the stock price…
…You don’t have to adopt David’s rule as a rigid requirement. There are legitimate reasons to sell: A business may suffer permanent deterioration. Your original thesis may be proven wrong. Management may stray in ways that betray your trust…
…The art of not selling isn’t really about not selling. It’s about becoming the kind of investor who doesn’t need to react to every bit of news.
2. Ergodicity and Investing – Eugene Ng
The average investor made money. The average investor also does not exist. There is no average investor. There is only you, your portfolio, your decisions, and your one path through time. Finance forgot that. Ergodicity remembers it…
…A system is ergodic if the average outcome over many people (ensemble average) equals the average outcome of a single person over time (time average). When those two diverge, the system is non-ergodic, because you are not the group.
Imagine 100 people each play Russian Roulette once. One bullet, six chambers in a pistol, spin the chamber, and fire the pistol. Survivors get a huge prize. The group average survival rate looks seemingly acceptable (83% = 1 – 1/6). That is the ensemble average. The expected value is 0.833 (5/6 x 1 + 1/6 x 0). A classical economist would say, positive expected value, rational to play. If the prize is $1 mil, $10 mil, or $100 mil, does the size of the prize matter? Would you still play such a game?
Now, imagine that one person can only play 100 rounds of Russian Roulette sequentially. They are dead with near certainty (~99.999999%). While in round 1, the probability of death is 16.7% (i.e., 1/6), which rapidly increases as more rounds are played. Probabilities grow rapidly to 60%, 84%, 97%, 99% after 5, 10, 20, 30 rounds, respectively. This is the time average.
It’s the same game, but over time results in a completely different outcome…
…Maximize growth that first conserves survival. Game-overs cause non-ergodicity. Do not maximise growth over survival. When permanent game-overs are possible, don’t rely on averages. Focus on not being wiped out permanently first.
Avoid a total loss and irreversibility at all costs. Never allow a single negative event to maximise short-term returns, rendering long-term maximisation irrelevant. If you are going to play a game where, after many rounds, you are almost certainly going to be dead. Avoid playing all games that are not repeatable at infinity…
…Survival beats performance. Performance is always subordinate to survival. The longer the time horizon, the more true this becomes. To be among the best over time, you need to keep playing the game, rather than being kicked out.
Focus on being antifragile, not fragile. To determine whether something is fragile or antifragile, expose it to volatility and see how it responds. Fragile things are harmed by volatility, and antifragile things benefit from volatility. Fragility is non-ergodic. Antifragility is ergodic. Fragility has limited upside and unlimited downside. Antifragility has a limited downside and unlimited upside…
…We avoid margin/leverage at all costs. Brokers can offer up to 100x leverage, but never take it. A 1% move against you could wipe you out. A Monte Carlo simulation of 20 sequential scenarios with 20X leverage, using 8% p.a. returns and 18% annualized volatility, shows that ~90-100% of the time, one will eventually be permanently wiped out (with a cumulative loss of -5%)…
…Don’t agree with redistribution, particularly for investing. Trimming your winners to feed your losers is incorrect, as it assumes the same likelihood of returns. Winners tend to keep winning, and losers tend to keep losing. Persistence tends to be more likely at both the right tails (winners) and left tails (losers). As long as the risk is overly significant, one should first let your winners run high, second, don’t trim them, and third, add to them.
3. Good news: AI Will Eat Application Software – Alex Immerman and Santiago Rodriguez
Yes, AI is a big deal. But the conclusion that AI is going to kill the vertical and functional software business model simply makes no sense. The truth is that AI simply isn’t going to kill software companies: after all this panic has passed, we’ll see that AI is the best thing that ever happened to the software industry…
..The bear case rests on a basic misunderstanding of what software companies actually sell. The market is treating “software” as though it were a commodity input—as if the value of a software company resided in its code, and cheaper code meant more competition and therefore cheaper companies. But code is never where the value has lived: if code is where the value was, these companies would have never gotten so big in the first place. They would have been killed years ago by open-source software or by competition from cheap software engineering labor in developing countries…
…AI might increase competition; but it’ll also dramatically expand what software companies can do, how fast they can do it, and how large the markets they serve can become. The end result won’t be margin compression to zero. Software will be a much bigger industry, with durable competitive advantages for the companies that earn them…
…The classic contemporary book on business moats is Hamilton Helmer’s Seven Powers. He lists seven distinct ways in which companies develop robust competitive advantages: Scale, network effects, counterpositioning, switching costs, brand, cornered resources, and process power…
…Switching costs are perhaps the one moat that really is going to change. It’s definitely true that AI is changing the friction and the cost-benefit analysis associated with switching vendors: agents can assist with a lot of migration work that used to be a headache…
…Network effects are a classic moat. And they aren’t going away…
…On the surface, Salesforce is a CRM database; but anyone who has worked in an enterprise setting knows that Salesforce is also an ecosystem. When everyone uses one platform, the network becomes self-reinforcing: you use Salesforce because everyone uses Salesforce. And the more companies use Salesforce, the more valuable the ecosystem of third party applications built on top of Salesforce and platform administrators experts in Salesforce…
…Scale was never the defining moat in software—it’s just not as important for Salesforce as it is for a cloud provider or for an industrial company. But to some extent, it may matter more for AI applications where compute spend exceeds labor costs, driving a unit cost advantage to the larger consumers of tokens. In addition, there are places where scale will still help: it’s a straightforward economy of scale to concentrate that maintenance burden in one place, since productivity gains from specialization don’t go away in an AI world…
…Cornered resources, like high-quality proprietary data, aren’t going to stop mattering either. If friction goes to zero, simply consolidating publicly available data into a usable interface becomes less valuable, because anyone can do it. But if AI enables doing much more with high-quality data than you could before, then the stuff that you can’t get easily becomes extremely valuable…
…And perhaps the strongest moat of all in this new era is process power—or as George Sivulka of Hebbia calls it, “process engineering.” Application software can be thought of as a stored process—it encodes opinions about how the function of an organization should operate, and those opinions calcify over years and decades of use into something that is inseparable from the organization itself. Successful app software companies are the ones that co-evolve with their clients around these workflows. As those workflows penetrate ever-deeper into an organization, process engineering only becomes more important. And more difficult for challengers to replicate…
…Counterpositioning is a kind of power that can be summoned and wielded by new entrants to a market. It’s when the new company has a business model which, for whatever reason, is unattractive for the incumbent company to compete against. Disruption theory from Clay Christensen is a classic type of counterpositioning, but it doesn’t always have to be “low cost” as the differentiated counterposition. In software, a new technology stack could create the opening for a startup to create new kinds of products and business models that are difficult for incumbents to replicate – like Databricks and their “Lakehouse” model.
The “agent” model of doing work and replacing tasks is certainly going to create some counterposition opportunities for new startups to challenge incumbents. There’s been a lot of ink spilled on the disruption of “per seat pricing” at the hands of agentic upstarts with value-based pricing. Let’s take customer service as an example. Decagon prices its customer support product per conversation handled, not per agent seat, and will eventually price per resolution achieved: that’s fundamentally a better alignment of incentives between vendor and buyer. An incumbent like Zendesk can’t easily make that same move without cannibalizing its own seat-based revenue. Just as Blockbuster couldn’t match Netflix’s subscription model without destroying its existing economics or Peoplesoft couldn’t match Workday’s SaaS model without upending its monetization. Companies that start with the new business model don’t face that dilemma, and it’s the core reason why platform shifts so reliably produce new winners.
But guess what? The total amount of “end state pricing power” in the market didn’t necessarily decrease; it just means customers now have a choice of business models they’d like to subscribe to, and the better one will win. That’s how competitive markets have always worked! AI is not the first time that a wave of creative destruction has rearranged markets and shifted the playing field. But here’s the thing: the business models that result almost always dwarf the old ones in the scale of the total opportunity…
…AI isn’t going to destroy the software industry; it’s going to split it into two parts. There really will be some categories of software companies that face genuine pressure. Frontend tools that serve primarily as thin wrappers around commodity functionality and do relatively little beyond presenting data in a slightly more convenient format are vulnerable. Incumbent systems of record that still operate on archaic interfaces but raise prices every year should be worried. So should software companies that have an outdated pricing model and value proposition that’s just inferior to what AI-native competitors can offer. The companies that win in this environment will be the ones delivering genuine value, not the ones that built the highest walls around their customer base.
4. THE NDFI BOMB – Dirt Cheap Banks
Here is a sentence that should terrify you: the single fastest-growing loan category in American banking is one that most investors have never heard of, most analysts don’t understand, and most banks can’t fully explain.
That category is loans to Non-Depository Financial Institutions, or NDFIs.
An NDFI is any financial company that lends money but doesn’t take deposits. Think mortgage companies, private equity funds, hedge funds, subprime auto lenders, fintech lenders, insurance companies, business development companies (BDCs), and the sprawling private credit universe. These are the shadow banks. The firms that exist in the regulatory gray zone between Wall Street and Main Street.
Here’s where it gets dangerous: traditional banks are funding the shadow banks. When Bank of America extends a $500 million credit line to a private credit fund, or when a regional bank in Indiana provides warehouse lines to a dozen mortgage companies, those are NDFI loans. The bank is one step removed from the actual borrower, lending to the lender, and often has limited visibility into what’s happening with the money downstream.
As of Q1 2025, U.S. banks held $1.14 trillion in outstanding NDFI loans, according to the Federal Reserve Bank of St. Louis. But that’s only the money that’s already been lent. The International Monetary Fund estimates banks have an additional $900+ billion in undrawn credit commitments to NDFIs. That’s money NDFIs can draw down at any time, for any reason. In a crisis, they will.
Total potential exposure: north of $2 trillion.
And it’s growing at a pace that should make every risk manager in America lose sleep. NDFI lending has grown at approximately 26% annually since 2012, according to the St. Louis Fed. In 2025, it surged more than 50% year-over-year according to Federal Reserve data, the largest jump in records going back to 2016.
To put that in context: total bank loans grew roughly 4% annually over the same period. NDFI lending has been growing at six times the rate of everything else…
…The FDIC now requires banks with over $10 billion in assets to break their NDFI lending into five categories. Here is where the $1.14 trillion actually goes, based on Q4 2024 call report data:
Mortgage Credit Intermediaries (23% of all NDFI loans, roughly $219 billion): These are loans to non-bank mortgage companies. The bank provides a “warehouse line” that the mortgage company uses to fund home loans. Once the mortgage is originated, the mortgage company sells it to Fannie Mae, Freddie Mac, or Ginnie Mae and pays back the warehouse line. The end-borrower is a homebuyer. The risk to the bank is that the mortgage company goes bust before it can sell the loans, or that the loans don’t qualify for agency purchase and the collateral is worth less than the advance. This is generally considered the lowest-risk form of NDFI lending because the collateral is agency-eligible mortgages with a ready secondary market.
Private Credit Intermediaries (23%, roughly $202 billion in private equity fund loans plus additional business credit intermediary exposure): These are loans to private credit funds, business development companies, and leveraged lending vehicles. The bank provides subscription lines (backed by investor capital commitments), NAV facilities (backed by the fund’s loan portfolio), or direct credit lines. The end-borrowers are mid-market and lower-middle-market companies, often highly leveraged, that couldn’t get financing from traditional bank channels. These companies typically carry 4x to 6x debt-to-EBITDA, and in some cases higher. The bank’s collateral is ultimately the fund’s portfolio of leveraged loans to these companies.
Business Credit Intermediaries (21%): Loans to companies that in turn provide business financing. This includes BDCs, equipment leasing companies, specialty finance firms, and factoring companies. The end-borrowers are small and medium businesses.
Consumer Credit Intermediaries (9%): Loans to non-bank consumer lenders. This is where subprime auto lending lives. Tricolor Holdings, the company whose collapse kicked off the NDFI panic in September 2025, was a consumer credit intermediary. It sold cars and provided financing largely to borrowers with little credit history. JPMorgan, Fifth Third, and Barclays all had warehouse-style exposure. The end-borrowers are consumers who can’t qualify for traditional bank financing.
Other NDFIs (24%, roughly $395 billion): A catch-all category that includes insurance companies, pension funds, broker-dealers, investment banks, bank holding companies, and securitization vehicles. JPMorgan classified its entire $133 billion NDFI book as “other”, declining to break out subcategories, citing “organizational risk” associated with reporting different values to the FDIC and the Fed, according to the Financial Times.
The bottom line: 46% of all bank NDFI loans fund mortgage origination and private credit lending. The end-borrowers are homebuyers on one side and highly leveraged companies on the other. The remaining 54% funds everything from subprime auto loans to hedge fund margin lending to insurance company investment portfolios.
5. All of the Jobs That No Longer Exist – Ben Carlson
Heading into the 19th century, about 70-80% of all jobs in the industrial world were in agriculture.
Most people were farmers.
By 1870, more than half of all men owned or performed labor on farms.
Today less than 1% of the U.S. population works in agriculture…
…There are plenty of jobs over the years that have been taken out by technology…
…There used to be people who would light all of the gas lanterns on the street by hand. They were replaced by electricity.
Before alarm clocks, people called knocker-ups used to go around tapping windows to wake people up…
…Before computers were around, NASA used human computers who literally did calculations by hand…
…It used to be someone’s job to set up the bowling pins by hand…
…There used to be video store clerks who would be forced to rewind the videos you forgot to rewind (and charge you for their troubles).
I could continue.
All of this job displacement and more has occurred yet the unemployment rate over the past 80 years or so has averaged less than 6%…
…There will certainly be a painful transition for many white-collar roles as AI is integrated into the workflow. I’m sure there are jobs out there that will be impacted by AI that we’re not even considering right now.
But new roles will also be created. AI will make so many people better at their current roles. That’s going to lead to more opportunities.
For many workers and businesses, AI will lead to more customers. Lawyers will be able to file more lawsuits. Tax accountants will be able to file more taxes. Financial advisors will be able to handle more clients. When bottlenecks are removed, output increases.
Disclaimer: None of the information or analysis presented is intended to form the basis for any offer or recommendation. We currently have a vested interest in Netflix and Salesforce. Holdings are subject to change at any time.
