We’ve constantly been sharing a list of our recent reads in our weekly emails for The Good Investors.
Do subscribe for our weekly updates through the orange box in the blog (it’s on the side if you’re using a computer, and all the way at the bottom if you’re using mobile) – it’s free!
But since our readership-audience for The Good Investors is wider than our subscriber base, we think sharing the reading list regularly on the blog itself can benefit even more people. The articles we share touch on a wide range of topics, including investing, business, and the world in general.
Here are the articles for the week ending 20 September 2026:
1. Fences, not Sandboxes – Steve Yegge
I’m here to give you a glimpse of a future that I think none of us expected. It’s a future where AIs are governed by laws, not by programs that try to contain and control them…
…My current software factory, the Wheelhouse, is not for navel-gazing: I built it specifically to work on my game, Wyvern. For all the skeptics out there saying, “Where’s the thing people are building,” well, I’ve got mine…
…Long story short, Wheelhouse was built via me complaining endlessly to Fable about what I want out of Wheelhouse — mostly more code launched, faster, but also lots of bespoke monitoring.
I would also notice when Fable would go off the rails, and gently nudge it back. I would let it fail for days to weeks, then make it do things my way. Fable is extremely data-driven, if you permit it to be, and it will insist on experiments and quantitative validation of anything you try to change. But the numbers would almost always prove me right, and Wheelhouse has been in a state of constant innovation.
But that innovation is all directed towards Wyvern. Wheelhouse exists to build and operate Wyvern; it has no other purpose in life. And yet in ten weeks, it has grown from nothing to rivaling the size of Wyvern itself. Wheelhouse is about 600k lines of code and tests (mostly bash), and Wyvern’s code (not counting content) is only about twice that big.
So the factory for building Wyvern is growing much faster than Wyvern is — even with heavy brakes applied lately, after Sol told us to tighten it the F up in a code review. We avoid new machinery but it still continues to grow rapidly, and I’m honestly not sure what the ideal factory-to-product ratio is yet. But it seems to be approaching 1:1.
You might wonder if Wheelhouse is reusable code, whether I could open the repo and let people try it out. I had no idea. I knew that my agents had built something really powerful, capable of shoving 500 commits per day through our merge queue (though we average 270/day), using magic tricks that are a year ahead of their time. It’s a system that we can ride so hard that it scares the players and they tell us to slow down.
But I wasn’t sure if it was reusable. I wasn’t even sure how it worked.
My agents had been using a lot of jargon, and I slowly realized they were reusing the same terms, day in and day out. They were speaking about things in Wheelhouse, using what seemed like recurring new design patterns: fences, ratchets, governors, tripwires, latches, gates, falsifiers… it was a long list, but finite. I just had no idea what any of these jargon terms meant.
So one day, no more than a week ago, after the 100th “fence” reference, I decided to peek under the covers and see exactly what my Fable agents had built. I had them create manifests, taxonomies, audits, and visualizations. They showed me what they had wrought.
This is the part where words fail me and the blog just falls over. My reaction was straight up WTF. No words.
Because I expected them to have built an engineering system. One that, you know, does stuff.
Instead, what they had built was an entire legal system, complete with a constitution, jurisprudence, courts, offices, jurisdiction, case law, rulings, registries, ledgers, rosters, and a full-fledged apparatus for running something resembling a manorial estate.
In short, Fable had produced a medieval government. And there’s no doubt that it was heavily influenced by the target product, Wyvern, which is a medieval fantasy RPG, at least in the fanciful naming we used: Marshal, Seneschal, Reeve, Beadle, Portcullis, etc. But that LARPing was masking a bona-fide system of constitutional governance.
Wheelhouse’s legal system also has an enforcement arm. The fences, gates, ratchets, and so on — when my agents used that jargon, they were referring to the enforcement machinery: the cops, as it were. And cameras, and jails…
…I’m here to tell you that if you allow it, Fable will try to capture all of that into a mechanically provable, AI-operable model of your organization, one where there are no unwritten rules. If there is one unwritten rule in Wheelhouse, it’s that the system hates unwritten rules.
Fable will capture all your rules, and write them down, if you let it. Then it will try building infrastructure to help enforce them.
I see it happening already, and people are fighting it. I see people making Skills to keep Fable from building “extra” stuff. But all Fable is trying to do here is the Right Thing. And that starts by capturing how your system operates, and how it is intended to operate, so it can begin addressing the gaps…
…When you add it all up, Fable is trying to turn Wheelhouse into an engine that can prove, mechanically, that every change to Wyvern is legal. The agents capture every single intention, decision, policy, rule of thumb, and legacy behavior in the system, and they use that to govern every future decision and action. They live by the Rule of Law.
Did they do a good job of all this? Well I mean, for sixth graders, yes, it was a great project. Once I popped the hood, I saw that they hadn’t been curating it, just growing it. It had a lot of cruft — for instance, old rulings that were obsolete or had changed. And ‘rulings’ that turned out to be just good craftsmanship, so we elided them. Like any engineering project, it needed ongoing maintenance.
I minted a new Officer seat, Frog (Head of Wheelhouse Law), and put Frog to work on folding successive cancelled rulings, and a whole bunch of other stuff the agents had overlooked. It’s a work in progress.
But on the whole, it was already a pretty solid system. The garden needed a bit of pruning and weeding, but not a redesign. Which is good, because redesigns are slow. Wheelhouse has a whole system just for the lifecycle of rules/laws: proposing, evaluating, ratifying, enacting, enforcing, measuring, amending, and retiring them.
And Wheelhouse is exceptionally careful not to break itself. So I can’t just make changes to Wheelhouse; they have to go through a ratification and review process, and then a build process, before they can take effect and propagate.
It’s running smoothly, though there are still all sorts of problems at this velocity. At hundreds of commits per day on master, idleness means staleness, and clones can fall far behind if they aren’t regularly pulling while they work. It takes external forces to get this to run reliably, so Wheelhouse has various roles for poking and prodding other agents.
In a lot of ways, it’s just like any other software factory.
The difference is, Wheelhouse is governed by a constitution. Humanity has only one mature technology for coordinating mortal, replaceable strangers via text — namely, law. Wheelhouse has 50 agents that are amnesiac and interchangeable, and the only way they can coordinate is via text. So offices outlive their holders, precedents outlive their incidents, and jurisdiction says who may act. Every group of cooperating humans eventually arrives at a system of laws, and agents are trying to do exactly the same.
So that’s the future. Hundreds to thousands of AI employees at every company, together comprising a city that needs an entirely new bespoke set of laws and rules.
2. First Impression of Muse – Abdullah Al-Rezwan
Before Muse, I poked around a bit with Instinct, a startup that has created a bit of buzz in the personal agent space but still in private beta. My very first impression of using Muse was that it seemed a pretty good clone of Instinct. But once I spent more time on Muse, I thought labeling Muse as a clone of Instinct is a very uncharitable framing.
Like Instinct, you can use Muse on WhatsApp (I didn’t see iMessage integration which you can do with Instinct). But Muse also has its own app which I installed to play around throughout the day. While the WhatsApp version seems just as barebone as Instinct, the app is much more feature rich. The app is available only in the US so far. Given ~40% of my subscribers are from outside the US, I wanted to give you a better feel about the app just by going through my own experience with ample screenshots along the way.
Muse has a very generous free tier. Zuckerberg mentioned in an interview yesterday that the free tier has up to a 100 million weekly token usage limit. While that sounded plenty to me at first, as of this writing, I have already used 81% of my free plan’s weekly limit. I have an option to upgrade to two plans: a) Power ($20/month with 500 million weekly Muse tokens), or b) Maximum ($100/month with 3 Billion weekly Muse tokens)…
…The best use case that I have found so far is I can finally connect all my Gmail accounts with Muse. While using Claude or ChatGPT, one of my persistent problems has been that I can only connect one Gmail account (for what it’s worth, Instinct can do this too). Unfortunately, I have four separate Gmail accounts (one for personal, one for business, one from Cornell, and one that I randomly opened and very sporadically use). The problem is I have used three of these Gmail accounts in different settings and given Gmail’s own poor search functionality, I actually need an AI to go through all my Gmail accounts simultaneously to find a specific email I am looking for…
…I also liked Muse’s “Connectors” which let me connect a bunch of apps from my phone. Once connected, it’s a lot faster for Muse to do the work that needs the data from the app. As I will show later, even if you don’t connect the app, Muse can manually work in its own browser to try to complete a task…
…Let me start with booking accommodation, which has been pretty much every consumer chat bot’s highlighted use case, including Meta. So, I asked Muse to look for an Airbnb for a family trip (8 adults) near Mendocino from December 26 to 31 this year. I was just testing Muse’s capability, so an actual query would require me to share much more contexts than I did. Anyways, Muse used its browser functionality and basically approached it the way any human would: go to Airbnb, fill in the destination, dates, number of guests etc. and go through the search results. After it went through search results, it showed me a wall of texts with different options which you can see below. Brian Chesky has been quite adamant that text based interaction will not work very well for travel booking and a more visual search is required to bridge the gap between hype and actual consumer behavior. When I went through Muse’s suggestions, there are two ways I could go about it: a) just accept whatever Muse is suggesting, and b) actually click through each of the suggested options and carefully assess which one I actually prefer. I’m skeptical that most people will choose the former, and if you choose the latter, are you really saving much time here instead of just going directly to the Airbnb app itself?…
…There is, however, a tangible risk for Airbnb or any OTA out there. I could ask Muse to find out whether the Airbnb it picked is available at a lower price anywhere else on the internet. Muse tried, saw two options, but eventually found those listings to be inactive on those platforms. Airbnb may look fine in this example, but obviously there are indeed plenty of inventory that is multi-listed on different platforms or their own websites. In fact, I don’t have to manually ask this to Muse every time I book an Airbnb. I can just ask Muse to look into any accommodation or flight I book and always first figure out whether there is a cheaper way to book it. Once it’s saved in Muse’s memory, Muse can automatically do this without ever being asked again…
…Then for dinner, I wanted to see if Muse is up to the task of ordering via DoorDash. I asked it to show me some healthy options from Chipotle which I wanted to pick up myself…
…After some back and forth, I picked the chicken+ black bean bowl. To be clear, at least half the time I typically don’t know what I would like to order from DoorDash. I usually browse the app and it is through the browsing experience, I typically end up deciding what to order. So, this is far from my typical way of ordering food. I was just trying to test what Muse can do with a relatively simple food ordering query. Anyways, after picking an option Muse offered, I was prompted to share my DoorDash login credentials. I asked it to use my Gmail account to login. It tried but it got stuck in log-in circular hell. It would try to login, Google would let me know someone’s trying to login to my DoorDash account, I would confirm “it’s me”, and the whole process repeated three times after which I gave up.
For what it’s worth, I personally found these issues prevalent even while using Instinct. I tried the same Airbnb queries that I did with Muse. It took literally an hour, I kid you not, for Instinct to respond to my aforementioned Airbnb query. When I tried the DoorDash query on Instinct, it still took 8 minutes to respond. After experiencing Muse’s speed, the bar just became too high for a startup such as Instinct which probably cannot afford as much compute as Meta is throwing at these problems.
As you can see, there are still plenty of issues Meta (or any consumer agent) needs to work on to make it a very seamless experience. While it’s hard for me to see myself using Muse at the expense of apps such as Airbnb, DoorDash, or Amazon, it’s simply too early to either write Muse off as another flailing attempt at owning the consumer agent layer or assume Muse’s victory given its ample compute budget. I do sense a lot of switching costs though once I connect all my accounts and share login credentials with one agent. I would be very reluctant to repeat this whole process with another agent without a materially compelling incremental benefit.
3. A Stealth Startup Thinks It Just Hacked the Memory Shortage – Lauren Goode
Kepler Computing, a San Jose, California–based startup founded in 2018 by a team of physicists and computer scientists, says it has developed a new architecture for high-bandwidth memory (HBM) that directly addresses some of the chip supply bottlenecks that are constraining the computing market.
While chipmakers typically rely on expensive extreme ultraviolet lithography (EUV) to shrink the transistors on a chip, thereby packing more technology into the same amount of space, Kepler claims that its “3D stacking” approach and a proprietary new material allow it to increase density without relying on EUV at all—and it can work with existing semiconductor fabrication plants.
Kepler says it has made similar gains for the high-speed cache memory typically used in CPUs, GPUs, and XPUs. This so-called SRAM sits within the core of a chip die in order to cut down on data transfer times. HBM, by contrast, uses stacks of DRAM, which is a separate memory component of chips…
…For now, much of Kepler’s testing is happening in Singapore, which is where GlobalFoundries—a manufacturing partner and investor of $50 million—has a facility. Over the past two years Kepler has been building out what it refers to as “mini fabs,” where it produces its memory chips in conjunction with Global Foundries’ 28-nanometer chips. (It has also been running tests in GlobalFoundries’ facilities in Burlington, Vermont.)…
…In bypassing ultraviolet lithography, Kepler is one of a few tech startups working to avoid one of the major bottlenecks in semiconductor manufacturing. The startup claims it can produce SRAM that achieves the same density as 2-nanometer or 3-nanometer chips without having to invest in EUV…
…Kepler’s overarching pitch is that the accelerated computing market shouldn’t have to wait for brand-new memory fabs to be built in order to meet demand. Instead, new approaches to building memory within existing fabs can increase supply.
Their approach is twofold. In terms of improving HBM, Kepler says it has developed a novel 3D-manufacturing technique that fits more memory chips within a fixed footprint. The core compute can then sit closer to the memory, so the data travel between the two uses less energy. Their ultimate goal is to move data around in HBM with the amount of energy that’s comparable to SRAM, all while keeping HBM’s large capacity.
Kepler also says it has improved the density of SRAM using ferroelectrics, which can read and write data at lower voltages than the mechanisms typically used to process and store data in semiconductors. The startup did this by developing a new, low-voltage, composite material that works with this ferroelectric approach…
…In its early build-outs with GlobalFoundries, Kepler says it was able to convert a fab into a “next-generation” fab in just eight months, compared to a typical 24-month timeframe…
…Basically Kepler is betting that any additional costs that come from new materials or retooling existing fabs would far outweigh the $20 billion to $40 billion it costs to build new ones and outfit them with equipment worth hundreds of millions of dollars…
…Kepler Computing still has a long road ahead before it reaches full-scale production—assuming it gets there. To date, the company has run its technology on around 2,000 wafers. The startup says it’s planning to ship its first samples of HBM chips later this year, ramp up production out of Singapore next year, and start chip production in the US in 2028.
4. Software is about to eat the world much faster – Marc Andreessen
Software has been eating the world at the speed of human hands. It is about to eat the world at the speed of compute…
…In the past year, Devin has gone from writing 13% of Cognition’s production code to more than 90%.
When agents write 90% of the code, engineers can literally do ten times as much. The 10x engineer becomes the 100x engineer, as they shift from writing artisanal code to operating as the CTO of a fleet of agents. The 1000x engineer isn’t too far behind. As Scott puts it: “Within our lifetime, engineers will go from bricklayers to architects, focusing on the creativity of designing systems rather than the manual labor of putting them together.”
At Mercedes Benz, engineers turned what would have been an eight month long COBOL migration into 8 days of work with Devin. Rivian teams increased their test generation velocity by 10x. Devin triages and patches vulnerabilities across thousands of repos at some of the world’s largest financial institutions like Itau, where 70% of security vulnerabilities are automatically remediated by Devin.
Every time programmers get more leverage, doomers predict the end of software engineering. Compilers were supposed to shrink the profession. So was open source. So was the cloud. Instead, each leap made software cheaper to build, and demand for software, and engineers, exploded. It keeps happening because, as Milton Friedman observed, human wants and needs are infinite, so economic demand is infinite, and job growth can continue forever.
5. The turbulent AI era is here. The choices we make now are critical – Bill Gates
AI for the first time can replace and even exceed human cognition.
In terms of equity, AI will either be the greatest equalizer ever invented, or the worst source of injustice….
…Unfortunately, right now we are not preparing for it. I don’t see evidence that leaders, experts, and communities are confronting the challenges adequately. There is no plan to ease the entry into the AI era.
Part of the reason for this is that many commentators underestimate the extent of the impact AI will have…
…Another reason people underestimate AI is that analogies to the effects of past innovations are misleading. We have no experience with a technology that can be adopted quickly or that can think and move like a human. When the PC came along, it took twenty years to significantly change how we worked because the software had to be developed, the price had to come down, and people had to learn how to use the tools and incorporate them into their business processes. AI, on the other hand, runs on the devices we already have, and it uses natural language. We don’t have to adapt to it because it can adapt to us…
…The transition to AI comes with three big risks…
…Many jobs will disappear forever.
In 1933, during the Great Depression, unemployment in the United States was roughly 25 percent. It remained in double digits for much of the following decade. It ultimately recovered as demand, investment, and growth returned.
AI may not reach this level, but its impact will not go away with an economic cycle. The jobs at most risk are entry- and mid-level, and the new jobs being created will mostly require skills that take many years to learn.
White-collar jobs are already being hit modestly. After the widespread adoption of generative AI, employment fell significantly among young workers in jobs that are especially vulnerable to replacement, but not among their older colleagues.
I think this trend will continue, but it will not be confined to a handful of industries or occupations. Jobs in sales and customer support (online and over the phone), software engineering, and paralegal work may be among the first affected, but the disruption will reach much further as AI takes on tasks that today still require trained workers: things like assessing loan applications, doing data analysis, and even triaging patients. A few areas like software engineering will generate new demand as the costs go down, so the net job loss in those areas will be less than in others as long as some tasks, such as design, are better done by humans.
Blue-collar jobs will be affected as well. Although robots are not as far along as AI, eventually their cost will be dramatically lower too…
…AI will empower people (and perhaps AIs) to do more harm.
Long before AI entered the mainstream, there was information online about how to create weapons like bombs, bioweapons, even computer viruses. AI will make it much easier to not only get this information but act on it. Even criminals with very limited skills will be able to target victims at every scale: individuals, companies, and governments…
…AI capabilities are starting to be used for cyberattacks. The smartest cybersecurity experts I know are scared about the next few years, because the attackers are getting powerful new capabilities faster than the defenders can fix all the weaknesses. After all, the same AI model that can find a flaw in software so a company can fix it can also help a criminal exploit it. The resources needed to make an attack are going down significantly and we haven’t been able to separate those abilities from benign usage…
…AI could stunt our kids’ development and replace human relationships.
When I was growing up in Seattle, I didn’t have that many friends aside from a few other boys who were like me. It took hard work and a lot of help from my mom to develop my social skills so I could relate to different kinds of people. I still draw on those lessons today at the age of 70.
I doubt I would have put in the same work if I had had an AI companion back then. They talk to you in ways you’re already comfortable with. They don’t push you outside your comfort zone. They are always available and never get mad at you. This gives them the potential to become highly addictive and to rob us of the lessons we learn from connecting with other people.
The body of evidence on this subject is still small and a bit mixed, but there are signs that we should be very concerned. For example, in one study of more than 1,100 people who use AI companions, researchers at Stanford and Carnegie Mellon found that those with smaller social networks were the most likely to turn to a chatbot for companionship. And the heavier and more emotionally personal that use became, the worse they felt.
Young people could be affected for their entire lives. In his book The Anxious Generation, Jonathan Haidt makes an observation about the effect of social media that is even more true for AI: “Like young trees exposed to wind, children who are routinely exposed to small risks grow up to become adults who can handle much larger risks without panicking. Conversely, children who are raised in a protected greenhouse sometimes become incapacitated by anxiety before they reach maturity.”
An AI companion designed to never upset you is a big, protected greenhouse…
…We need both: deep concern about the AI harms we need to minimize, and grounded optimism about the positives if we maximize them for everyone…
…With its ability to synthesize knowledge from every scientific field, AI can accelerate innovation in the world’s toughest technical challenges: providing reliable clean energy for everyone, combating climate change, growing enough food, eradicating diseases, and more. Researchers working on cancer treatments or nuclear energy can use AI to search through massive amounts of scientific literature. It can help them identify patterns that a human might miss and decide which experiments offer the most promise. When intelligence is no longer the limiting factor that it is today, smaller companies will be able to compete with organizations that have far larger research budgets. R&D and innovation will be supercharged.
Healthcare is one area where AI can help solve real-world problems. Many small American hospitals lack on-site specialists who can quickly diagnose a patient during a life-threatening emergency. In those places, AI could make sure a heart attack is caught in time and a family avoids the crushing expense of a medical emergency…
…I surprise a lot of people when I tell them that a second area—agriculture—is where I see the fastest impact of AI in low-income countries. In most low-income countries, farmers don’t get reliable weather forecasts or advice on what seeds to plant, how to protect their crops and livestock from disease, or how to improve their soil. With population growth in these countries and the challenges of climate change, these farmers need more help than ever. Using AI, low-income farmers will soon be able to get better advice about all these things than even the richest farmers get today and increase their output substantially…
…The highest priority is a monumental task: creating a domestic and international framework for dealing with AI.
None of our current institutions were designed to handle a technology that spreads so fast and touches so many parts of our lives. So we’ll need to make new ones.
It’s hard to overstate what an enormous undertaking this will be. After the attacks of 9/11, the U.S. government went through its biggest reorganization since World War II for the purpose of improving just one function, national security.
AI will require much, much more. It will affect national security as well as employment, education, taxation, energy, elections, air and water, public health, the financial system, law enforcement, transportation, public lands, and IT systems…
… I believe that as AI and robots improve, we’ll set aside certain things for only people to do. I’ve started calling this domain Human Reserved, and it’s an example of the kinds of ideas we’ll need to consider.
I like the phrase Human Reserved because it makes me think of nature reserves—places where we could put buildings and roads, but we choose not to because the loss would be too great.
We might set something aside as Human Reserved for economic reasons. For example, we may do it because allowing machines to take over a certain role will displace a large number of people who can’t easily change jobs. You can’t tell a 55-year-old who has worked in construction their whole career that they need to go work at an elder care facility and expect them to find it fulfilling.
Sometimes the decision to make something Human Reserved will be driven by other factors. In health, for example, imagine a robot giving you the awful news that you have an incurable disease. There’s no technical reason why it couldn’t. Yet it shouldn’t…
…I believe we should tax AI tokens and robots. Right now, if you’re an employer and you hire someone, you pay payroll taxes on their earnings. But if you buy a robot, you can usually write it off right away as a business expense. The tax system nudges you toward replacing people with machines.
A tax would slow the rush away from human labor a little and raise money for retraining and a stronger safety net. It would need to be targeted so it does not slow down the purely beneficial uses of AI, like making medicine and education cheaper.
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 Chipotle Mexican Grill and Meta Platforms. Holdings are subject to change at any time.