What previous capital cycles can teach us about how today’s extraordinary memory boom may eventually end.

Muse became the No. 1 app in Apple’s App Store this week, ahead of ChatGPT. Meta rallied sharply and Shopify jumped after announcing that Muse would be able to use Shop Pay to complete purchases. What interested me more was the conversation inside our community Slack. One member described Muse as the easiest introduction to agentic AI he had seen. Another has been using it to track down a discontinued stuffed animal for his stepdaughter, book lap swimming at his health club, check ferry times for an upcoming trip and monitor a futures watchlist before settlement. It just works. For much of the AI race, we have been obsessed with which company has the best model. Muse is interesting because the model underneath it is not obviously the best. Ben Thompson made the point that model capability may now be good enough for companies to build compelling products with actual moats. A personal agent also has the potential to become much stickier than a chatbot. Once it knows your email, calendar, preferences, payments and the hundreds of small things you have taught it about your life, changing agents becomes more difficult. The advantage starts to come from everything built around the model. You could see the market beginning to work through the implications this week. Shopify rallied after integrating Shop Pay with Muse. Amazon went the other way and blocked Muse from shopping on its site. Travel companies, insurers and other businesses perceived to benefit from consumer inertia came under pressure as investors wondered what happens when an agent can constantly compare prices and switch providers on your behalf. One of our members made a distinction I keep thinking about. The final winners may come down to distribution with trust and safety. That feels especially important for personal agents. The more useful they become, the more of our lives we have to let them into. Muse already had a security vulnerability discovered and patched this week. Convenience gets you surprisingly far. Trust may determine how far. There is another investment implication here. One of our members pointed out that the market has repeatedly assumed AI would disrupt the existing technology monopolies. Google was going to lose search. Microsoft was vulnerable in software. Apple was hopeless because of Siri. Meta was spending too much on AI. So far, the incumbents keep adapting. Perhaps we have been too quick to think of AI as something that destroys existing moats. There is a difference between a business protected because customers find it annoying to leave and one protected because customers already live there. AI agents may expose the first kind of moat while making the second considerably more valuable. I made the case in our salon yesterday that this could lead to another period of Mag 7 outperformance.

Of the 4 percentage point run-up in long-term Treasury yields between August 2020 and the start of this month, 90.5% took place in a three-day window around the monthly jobs report or major Fed speeches.

One of our members noticed that his worst losses tended to come after long stretches of being right. So whenever he feels completely certain about a position, he asks himself: How do you live with being wrong? A useful question to carry into a bull market.

I’m delighted to share that the Stray Reflections audiobook is now live. Something strange happens when you have more information. You don’t always become more certain. Sometimes you become less. More research means more evidence to weigh. More opinions mean more reasons to question your own. At some point, the problem stops being access to information. It becomes knowing what matters, and having the conviction to act on it. That’s what Stray Reflections is about: thinking clearly in a world with more noise than ever. And now, you can listen to the full book for free. stray-reflections.com/audiob…

Markets entered this year expecting rate cuts. Nine months later, the Fed is hiking. That’s a pretty extraordinary reversal in the macro backdrop, but what interests me is how well equities have absorbed it so far. All of this with the 10-year Treasury yield around 5 percent, oil still above $100 and the Fed signalling that another hike may be coming. If the Fed has to raise rates further, the conventional expectation is tighter financial conditions, weaker growth and eventually lower inflation and bond yields. Long-term yields, however, have already been remarkably resistant to the things that would normally push them lower. Earlier this year, falling oil, softer inflation and weaker employment failed to produce the bond rally many expected. One of our members made the point that despite a hawkish hike, the long end seemed more concerned with the wars and energy than the Fed. If that’s right, the Fed may have to tighten considerably more before it gets the response it wants. AI may be another reason. I wrote on Monday about the calls from Dario Amodei and Sam Altman to slow the pace of AI development. They can both genuinely want a safer pace and still wake up the next morning knowing the other is capable of making the next breakthrough. Normally, extraordinary investment eventually runs into the discipline of returns. If the returns aren’t good enough, companies spend less. But AI has another variable: the cost of falling behind. If the companies building frontier models believe the next breakthrough could materially change their competitive position, some of this spending starts to look defensive. Economics may take longer to stop the cycle than we expect. That matters when the Fed is trying to slow an economy in which the demand for capital remains so strong. We tend to think about a Fed hiking cycle as though the policy rate eventually overwhelms everything else. Raise rates enough and demand weakens, growth slows, inflation falls and bond yields follow. But what if it takes considerably more tightening to get there? The important question after this week’s meeting is how high yields have to go before something finally breaks. The macro surprise may be just how high that turns out to be.

A lot of my best clues this year have come from markets refusing to behave the way I expected. Oil wasn't high enough. Bonds wouldn't rally. Stocks wouldn't fall enough. I've become much more interested in failed relationships between markets. Sometimes the thing that doesn't happen tells you more than the thing that does.

Markets entered this year expecting rate cuts. Nine months later, the Fed is hiking. That’s a pretty extraordinary reversal in the macro backdrop. What interests me is how well equities have absorbed it so far.

The fear surrounding AI may itself extend the AI capital-spending cycle, because once falling behind becomes existential, investment is no longer governed entirely by conventional ROI.

Despite gold recovering less than half of its decline since the late-January peak, investors are already piling back into bullish options bets. GLD calls now exceed puts by 2.5 million contracts, close to the January 29 peak.

Farmers got expensive fertilizer during planting. Now they're getting expensive diesel during harvest. There isn't much discretion involved in either. The crop has to come out of the ground and then it has to be transported. That's one reason I'm paying much more attention to food inflation now.

One reason oil may have disappointed the bulls this year is that the world is simply better able to absorb an energy shock than it used to be. Economies use less energy per unit of output, supply is more diversified, and gasoline takes a smaller share of household income. Hormuz still matters enormously. Its economic leverage may just be different from the one investors learned in the 1970s.

I increasingly think OpenAI and Anthropic may tell us more about the eventual end of the AI infrastructure cycle than Nvidia. Nvidia tells us how much infrastructure is being bought. OpenAI and Anthropic are closer to telling us whether that infrastructure is earning an adequate return.

How the explanations we develop for who we are can quietly limit who we allow ourselves to become.

All you need to know about what’s going on with the US-Iran war.