Bits and Bobs 5/11/26
Alex Komoroske
Why is Claude Code closed source but Codex is substantially open source?
Claude Code was the breakout success of a new category that it created.
There was no reason to start being open.
That would be a cost for no benefit
Then there was never a time to open it and by the time it would have made sense to do it it was too late.
Anthropic got locked into the closed position, leaving OpenAi to counter position as open.
OpenAI has also made it clear you can use Codex and ChatGPT Pro in whatever harness you want.
Anthropic has, in a high-visibility way, come to exactly the opposite conclusion.
Doesn’t that lock OpenAI into an unsustainable position of subsidy?
Not necessarily!
Individual models often come down in cost by multiple orders of magnitude over years.
OpenAI can commit to this model being allowed to be used with the subsidy in perpetuity.
Over time that cost will collapse to be negligible.
But OpenAI still has the ability to not include future expensive models in that policy.
That gives them the benefit of unambiguously being a more open policy for the subsidy now, and with little downside in the future.
Competitors try to differentiate on salient dimensions.
They fall into a strategic position and then the world reacts and forces them to be more of their strategy to differentiate and counter-position.
Decisions of cosmic microdust blown up into whole strategic galaxies as the universe expands.
That means that if one successful entity has a closed version, another player is likely to produce an open version.
The entity that chooses to be open is typically not the one that has a good chance of being the market leader with a closed system.
So typically not the first place competitor.
Or even the second-place competitor (who hopes to become first place).
The third-or-below entity is way more likely to pick to be open.
Normal subsidies show users value so users continue to pay even when the subsidy tapers off.
But that requires the subsidy to be small percentages.
The subsidy for LLM tokens today is many orders of magnitude off, so it fundamentally distorts user behavior in ways that can’t be tapered off easily.
A betrayal: discovering an interaction you thought was with a human turns out to be a bot.
Especially if you feel like they’ve wasted your time.
The agent is not situated, it has infinite time.
Humans are situated, they have opportunity cost and finite time.
Due to that asymmetry, agents should never waste a human’s time.
For builders, agents are the ultimate slot machine.
They can often build what you ask for better than you could imagine.
Other times they fall on their face.
Variable random reward.
Keep pulling the lever and seeing if this next pull will make it work.
It very well could!
“Just one more pull…”
This is why a lot of builders are getting so little sleep in the last few months.
Exploring is all about surprise, while exploit is more about efficiency.
This is why the ROI of exploring is structurally harder to define ahead of time compared to exploit.
Exploit wins in organizations because it’s easier to make the ROI concrete ahead of time.
Exploit is fundamentally legible, explore is fundamentally illegible.
Saurmans are most at home in exploit.
Radagasts are most at home in explore.
Tokens are better spent on explore than exploit.
Exploring is about surprise, which means LLMs are great for it.
Exploiting starts off being about surprise, but as it gets dialed in on a given hill, becomes about efficiency.
Mechanistic software is orders of magnitude more efficient.
Luckily, LLMs can create mechanistic software for you!
LLMs can write mechanistic software well.
Tokenmaxxing at organizations is burning a lot of tokens for not much benefit.
Turns out if you only focus on token burn then you will burn a ton of tokens.
If you can do the same thing with more or less tokens, the one with fewer tokens is better.
If you’re tokenmaxxing, you'll use tokens for even the most mechanistic, repeated tasks.
The enlightened form of tokenmaxing is “How can I maximize my bang per buck?” not “How can I maximize how many bucks.”
A pattern for high-quality LLM code: in the design phase, tell the agent to make no assumptions without confirming.
The vast majority you’ll say “yup” to… but each one you tweak makes the downstream code generation significantly better and prevents tons of wasted effort.
This helps ensure its mental model is the same as yours.
It helps ensure fractal coherence of your plans.
If you can specify the code, AI can build it.
But be careful if your specifications don’t match your intentions!
And of course your specifications can’t match your intentions.
That’s why the monkey’s paw emerges.
The only reason specification works in the real world is because of non-transactional trust between situated humans.
LLMs will cut whatever corner they can if your spec allows it.
That’s why fractal coherence of your spec is important.
Agents are so effective that describing the problem to a coworker so they can help is often not worth it.
Before agents, even though it would take a ton of time to hand-off, it was still the only way to get leverage.
To distill it enough to hand it off to a human requires specifying in inverse proportion to the capability and alignment of the counterpart.
Hand-offs are also largely constrained by the patience and ability of the receiver to receive an infodump.
LLMs are extraordinary at receiving infodumps.
Orders of magnitude better than all but the most curious and motivated humans.
Giving the right context to LLMs is such a powerful unlock that it feels like cheating.
They’re a steam engine, just fill them with the right fuel and they can move mountains.
Apple Pulls ‘Anything’ App in Growing Clampdown on Vibe Coding Apps.
The app distribution model is a poor fit for infinite software.
One of the things that kicked off the era of Agentic Engineering was just robust file editing.
As recently as a few months ago, LLMs editing files was error-prone and slow.
Streaming out an edited file, token by token, and hoping the LLM doesn’t get confused or try to ‘improve’ something unrelated.
LLMs producing diffs would get subtly confused.
All of the agent harnesses do file editing in slightly different ways.
But they all give the LLM a mechanistic tool.
They typically require the LLM to provide the exact text it wants to modify, and treat multiple matches as an error.
Some also require the LLM to have read the file since it was last written before they allow a write.
If you turn these off, Agentic Engineering gets orders of magnitude less smooth.
One of those minor catalyzing innovations we take for granted!
Zeno’s paradox of infinite software.
The first 80% is insanely fast to create.
The last 20% to make it useful takes even longer than it used to.
Fit and finish is hard to do with a squirrely thing that’s hard to pin down.
The expectations are set based on the first 80%.
There is a vast amount of missing software.
LLMs are so insanely capable at producing code and so far the software we’re producing is just the same junk.
Tons of half-finished stuff like what we already have.
Where is the totally new, only-possible-with-LLMs software?
When it gets here, we might even recognize it as software.
Software that is infinitely malleable won’t look like software at all.
LLMs have only been getting incrementally better but new things are also made discontinuously possible.
That’s due to threshold effects.
Some use cases don’t work until a quality bar is cleared, then they just start working.
Before it required significant specification to get good results.
Now, you can get good results with minimal specification.
The models are just good enough to figure it out on their own.
Turns out LLMs are very good at learning the patterns underlying any sequence, no matter how subtle.
I love the frame in this article that OpenClaw is the Swedish Vasa ship.
It was so overstuffed and ill-considered that it capsized before even leaving the harbor.
The app model requires users to trust the software with their data.
That the creator isn’t malicious… or naive.
That is no longer a good assumption in the world of infinite software.
This week in the Wild West Roundup:
Someone Used Morse Code to Trick Grok Into Sending $174,000 and It Has Happened Before.
Vulnerability in Claude Extension for Chrome Exposes AI Agent to Takeover.
Microsoft: When prompts become shells: RCE vulnerabilities in AI agent frameworks.
Malicious OpenClaw DeepSeek Skill Exploits Agentic AI Workflows to Deliver RAT and Stealer.
An AI Agent Just Destroyed Our Production Data. It Confessed in Writing.
ClawHub has 44,000 skills. We audited 1,024. Here’s what we found.
Spoiler alert: it’s a mess.
One command turns any open-source repo into an AI agent backdoor.
Encoded Prompt Injection: Why LLM Guardrails Are at the Wrong Layer.
When you're using someone else's cloud, the only features you can have are the ones the cloud's owner decided to add.
You're on their turf, playing by their rules.
Roach motels are great businesses, but it's hard to get people to want to come in.
Only if it's so amazing that it's impossible to not go in will customers go in.
Once in, they are trusting that the roach motel owner won’t over-extract from them.
Good luck!
Over sufficient time, over-extraction from a roach motel owner is inevitable.
Folk Software: “Code that starts with "I wish this existed".
Even if you come up with a great way to use AI, it's hard to share it with another person for them to use.
This is in terms of both UX and security.
UX: your flow is highly bespoke to your particular set up.
Security: the code you share could be very dangerous.
When you’re riding a dragon, you can accomplish a lot.
You can create a lot of value… or do a lot of damage.
Riding dragons gives you leverage.
In large organizations, the faster you can sling shiny slop, the more you’ll get promoted.
YOLOing codeslop and workslop into the team’s commons.
When you accumulate Gilded Turds you’re left with a pile of shit.
Hoping to sling shiny slop faster than the others so you can get promoted and move on before the pile of shit becomes unignorable.
When you add slop to an incoherent thing, it becomes more incoherent.
This accelerates as more accumulates.
The way to clean up an under-maintained codebase is not to fling more slop into it.
A threshold: self-sustaining momentum.
Also known as PMF.
The value created is greater than the user’s expectations so they share with others, and draw in new users.
The more the value exceeds the expectations, the stronger the desire to share.
A linear increase in value to expectations has non-linear adoption effects.
In the early stage true momentum can be hard to observe.
For example, imagine looking at the stars on Github as a signal of momentum.
If there is a discontinuity, for example being on the homepage of HackerNews, that will lead to net new stars.
But only the stars produced by PMF really matter.
Those little blips of discontinuities obscure the underlying curve.
Some microbursts then lead to later aftershocks of other places covering it too, leading to net-new inbound.
But overtime all that matters is true PMF.
Default-converging things are fractally coherent.
At each level, nesting all the way down, there is alignment and coherence.
When you have fractal coherence, then more energy pumped in makes the thing stronger.
Things that don’t align perfectly are pulled into alignment by the gravity of the fractal coherence of the existing parts.
All of the neighboring parts are pushing in the same direction, not against.
The more that accumulates that is coherent, the more likely the next thing to be added will be made coherent, too.
When you don’t have fractal coherence, more energy pumped in makes it incoherent.
The more likely that things are incoherent, the more likely a new addition can’t tell what the proper alignment is and picks randomly.
The more incoherent things accumulated, the more likely new things are incoherent, too.
LLMs allow us to pump unlimited cognitive energy into projects.
Fractal coherence is even more important than ever before.
A continual process for managing a resonant spec for a product.
The resonant spec is all of the fractal clarity necessary for the thing to be coherent.
What it is and how it works and why it works, all increasingly aligned.
Keep a list of principles, from big to small.
Every principle that comes later in the list must nest underneath the ones that come earlier.
Look at the actions and decisions you or the system have made.
Keep the ones that were surprising at the time.
Systematize that list into your formal principles.
Ensure that everything below the principle in the list is in line with it.
If a given assertion appears to be load-bearing multiple times, promote it up the stack.
If lower-level assertions conflict, then adapt it to fit.
A sifting process for principles.
This can make sure the artifact is default-converging.
Useful for human teams… and working with LLMs too!
When the two ends of the arch touch, you get to a default converging state.
Before, people had to imagine the unbuilt parts of the arch.
It’s much easier for someone to get confused, especially if they aren’t already aligned.
When the arch is complete, but not necessarily strong, it gets discontinuously easy for others to participate and improve it autonomously.
It's obvious what it's supposed to do, and how to make it more robust.
That default alignment gets stronger and stronger the more robust and load-bearing it becomes.
It becomes a scaffolding for coherence.
Maggie Appleton in Zero Alignment:
"Anyone who’s shipped software on a team knows this isn’t a new problem. Alignment has always been a bottleneck.
But agents have made the cost of not being aligned as a team much higher."
The process of discovery is not linear.
The path will curve, sometimes wildly, from your start point to your destination.
But once you find the destination, you can head back to the start a little bit straighter.
Every time you walk the path, you cut the corners, optimizing the path just a bit.
Over time it becomes a straight path.
The same way that ants follow a trail.
Steady linear progress can lead to discontinuous results thanks to threshold effects.
Linear increases have no effect until there’s a discontinuous change because the threshold has been cleared.
Thresholds are often invisible before they activate.
The straw that broke the camel's back.
Enriching means you leave it better than you found it.
Do work that accretes so agents after you don’t have to do the full work.
This gives you compounding benefit.
Each little enrichment builds on the others, creating cumulative enrichment.
When two creative forces are engaged together, trusting one another, that dance creates a gravitational pull that pulls other energy in, too.
If their judgment and alignment with the goal is good, then it becomes default-converging.
Once it’s default-convergining, just run it hotter.
You can focus on getting teams working together well and aligned and then kind of sit back.
Claude is better at finding conceptual gaps.
Codex is better at finding implementation gaps.
Just because your code is mechanistic and non-fuzzy doesn’t mean that the problem it solves is not fuzzy.
It’s about fuzziness and uncertainty of inputs.
There are tolerances of software.
If you have 100 inputs that are exactly the same, and then 1 more thing that is wildly different, how much should you deoptimize?
How flexible your pipeline should be is similar to a multi-armed bandit problem.
Premature concretization is like having too many significant digits.
You can’t make something fundamentally non-concrete more concrete.
You can put spackle on a spring, but that won’t make it load bearing.
Excellent piece: Unraveling AI's 'Knitting Bullshit'
Bullshitting is hollowness even more than just lying.
Bullshitting is not caring about the truth.
The modern era is vibes without fundamentals.
Fundamentally hollow.
An infinite feed of Gilded Turds.
Adversarial systems help make your system stronger.
Is the goal of that adversarial system to benefit you, or to harm you?
You get stronger either way… but if the goal is to harm you, they might deploy the killing blow, whereas if their goal is to help you, you’ll definitely survive.
The beneficial adversarial approach is “white hat”.
The Anomaly Almanac: A meta-pattern for systematizing a real-world domain into a default-converging model.
Describe the model as it currently exists, as well as a list of “surprising” examples that don’t yet fit the model.
At the beginning, your model is empty, so every observation is surprising.
Then, fold the surprising examples into your model.
Your model should be adapted to work for all known surprising inputs in the past.
Now keep going, looking for more observations and keeping any surprising ones.
You can discard the ones that aren’t surprising.
Every so often, fold the new surprising ones into your model.
The list of at-some-point-surprising observations provides an easy test to check your model against as you build it.
How good your model is is how rarely you find surprising examples.
Of course, this presumes you are actively looking for disconfirming evidence.
That exploration will help you detect when the conditions have changed and you need to start smoothly adapting your model, instead of getting caught off guard and having your model shatter in a new context.
Once you have a few new surprising examples, you have a model that is “dialed in”.
This meta approach naturally samples from how common use cases are in the wild.
It starts off focused on whatever things you happened to sample at the beginning, but over time the surprising examples are inherently more fundamentally interesting.
This is the exact opposite of how surprises are treated in everyday execution in modern business.
Every surprise is inconvenient, possibly distracting, something to sand down and get rid of.
Instead, we should treat surprises as precious.
One watch out: be careful to get diverse-enough input.
We tend to only get input in domains where applying our model gives us good-enough results.
For example, if we build a product, the only users bothering to use it and give feedback are the ones who it was a good enough fit for already.
That can lead to pulling you at an accelerating rate into a small niche.
That’s why surprising examples are extraordinarily valuable.
Throughline Tacking: a meta-pattern for discovering the throughline of your product.
It can be hard to balance the messy reality of the world-as-it-exists and the world-as-we-think-it-ought-to-be.
It’s easy to go to either one extreme.
Here’s a meta-pattern for keeping both ends on track.
Like in sailing, you can sail into the headwind by tacking back and forth.
First, start with a messy-real-world thing that works.
For example, a product that has PMF.
Now, look at it and try to understand: why does it work?
Not “What do we want to work,” but “actually, what parts resonate?”
Actual products evolve from feedback and adaptation continuously.
They weren’t planned so much as grown.
But if you wanted to tell a convincing narrative about why it ended up where it is, what would it be?
This is a process of retconning.
The retcon is in some ways fake (that’s not why you built the product this way) but it is in a deeper sense real.
It is about the throughline of what makes it resonate in practice.
If the throughline points to a destination you want to go, great.
If not, then you can arc it to point in a slightly different direction in the long-term.
Now that you have documented the throughline, you can make the product more like the throughline.
Some features become more obviously vestigial or distractions.
Some minor features become much more obviously load-bearing.
Now, continue this process continuously.
By tacking back and forth you’re sailing along the throughline.
This process allows your product to become more what it wants to be.
This is a process for discovering and extruding emergent fractal coherence in a product.
This process works for any emergent artifact you care about, not just products.
Team culture.
Works of art.
Organizational bylaws.
LLMs plus full-genome sequencing could radically improve the diagnostic journey for rare conditions.
For example, this case.
It takes significant amounts of cognitive labor to sift through all of the symptoms and genetic markers.
LLMs are great at cognitive labor with infinite patience.
The medical system is tuned towards handling acute symptoms.
It is not well tuned to give advice like “Given you have this specific genetic condition, take this vitamin and avoid folic acid and you will have significantly less nerve pain in a decade.”
To diagnose a subtle problem takes tons of cognitive labor, much more than a doctor can do in a 15 minute visit.
Wellness is about being healthy, not just fixing the stuff that’s broken.
One reason osteopaths do a good business: people with a medical background who are willing to dig into a problem with a patient and do the cognitive and emotional labor.
A paper: O-Ring Automation.
Makes the case that estimates of labor displacement due to automation are structurally too high due to how labor is offset in real workflows.
Apparently colleges are trying out a new form of testing.
There’s a “Pupil bot” that pretends to not know the material, and the student has to explain it to them so the bot can pass the test.
Being able to explain it helps lock in your own understanding and is a better target than “do you know the answer” anyway.
The friction of contracting leads to hierarchy.
An alternate distillation of the Coasian Theory of the Firm.
Nimbleness and stability are fundamentally in tension.
At 0-to-1 companies, alignment and enthusiasm is so important that anything other than "Hell yes!!" is a “No.”
The candidate should be saying not “you guys” but “we” by the third interview.
The company should do an anti-sell at the end to make sure the candidate still wants to come even if they tell them all the reasons they shouldn’t.
That gives a gauntlet that makes sure only the most aligned and enthusiastic join.
When everyone on the team is aligned and enthusiastic, it is superlinearly more powerful than if only most people are aligned and enthusiastic.
If the person who pays your salary asks you for a faster horse, you give them a faster horse.
Only after giving them a faster horse and clearing that bar can you say, “I’ve been experimenting with this car thing…”
What is the order of magnitude time of the dominant coordination on a team?
Monthly cycle? Weekly? Daily? Hourly?
Different pace layers and different contexts demand different cycle times.
Founder mode values high autonomy in the team… as long as it’s aligned with the founder’s vision.
Even beneficial emergent ideas from the team are malicious emergence to the founder mindset.
The leader chooses the moon, the team chooses the roof shots.
Tightly aligned, loosely coupled.
Founder mode allows comfort with illegibility.
Managers must make it legible.
The founder is trusted to know what to do, it’s their thing they own, of course they understand it.
The manager is just an administrator; they must implicitly prove to everyone else that they are competent to run it.
That means managers must optimize not for correctness but legibility.
Founders don't need to optimize for legibility.
Over sufficient time, all niches saturate.
If you have the differential ability to succeed in a niche, you can go into it before anyone else bothers.
Then, by the time all of the generic low-hanging fruit are picked and your niche starts looking attractive to others, you already have a significant head start.
The late stage of any ecosystem: fighting harder and harder over smaller and smaller scraps.
Miserable for absolutely everyone, even the winners.
The only ones who are OK with it are the ones who got massive early and can now sit back watching as everyone else grinds themselves to a pulp.
Time for a new paradigm to emerge.
A 2x2 of different thinking styles.
Do you tend to over-think or under-think?
Do you tend to think broadly or deeply?
Which is more important to you:
Usefulness or elegance?
Consistency that is always good enough, or spiky quality?
Monkey’s paw is “[achieve this goal] at any cost”
It optimizes for the one dimension it’s graded on at the catastrophic cost to all others.
Are you asking questions to gain knowledge or are you trying to prove you're right and the other person is wrong?
It’s easy to get into a savior mode where you think you’re helping the other person avoid a problem via your questions.
But if you don’t actually understand the whole problem, this can lead you to be incurious about what you might be missing, and destroying value for the team.
The Product mindset is “what corners can I cut to get the user experience I want as quickly as possible.”
Life will find a way.
The LLM will find a way.
Any agentic goal seeking thing will find its way, given enough time and patience.
An individual might get bored and give up, but the swarm doesn’t.
Kayfabe will never go away as long as we have humans.
Humans who want to have an edge against their competitors for the scarce spoils will have an incentive to optimize for optics over fundamentals.
When time is scarce, optics matter more than fundamentals for how others will judge you.
The more cacophony there is, the more time scarce we become.
People have an incentive to make themselves indispensable by implementing a hacky approach that only they can do that’s load-bearing.
Eric Ries: “We’re in a war for the soul of our civilization between builders and extractors.”
Builders want to create value.
Extractors just want to extract value.
The current “best practices” of running companies are fundamentally tuned to extraction and thus are value destructive.
The easiest way to create short-term value is to light trust on fire.
The trust is invisible, so you won’t even see what you’re burning.
This is the way that many corrupt processes work; the individual benefits, the collective loses.
It’s invisible, so it keeps happening.
A 2x2 of incentives for an individual and the collective.
Aligned with the individuals’s incentives vs misaligned.
Aligned with the collective’s incentives vs misaligned.
Misaligned-individual / Misaligned-collective: Hell.
Misaligned-individual / Aligned-collective: Society grinds individuals down.
Aligned-individual / Misaligned-collective: Rewards corruption and destruction of value.
Aligned-individual / Aligned-collective: Transcendent.
The last is where selfishness creates net value for society.
Atelic and telic goals behave differently.
An atelic goal: being a musician.
You can never be done with that goal; it is an end in and of itself.
It’s inherently durable.
A telic goal: learning to play a given piece of music on a particular instrument.
It matters while you’re working on it.
Once you’re done, it matters much less.
A fascinating demonstration of the power of emergence: Simulating cells fighting to the death.
"Superficially, this is just a cute simulation of some artificial cells, but, for me, it gets at one of the deep wonders of life.
These cells exhibit high-level ‘purposeful’ behavior, but their motion is entirely driven by extremely simple low-level rules (and they’re not even deterministic!).
Just like in a real organism - just like in us, science believes - this purposeful behavior emerges from the interaction of many pseudorandom low-level components.
These cells were not programmed via top-down rules, as one might code a video game enemy (walk towards player, move limbs while doing so, fire when X feet away, etc.); their high-level behavior emerges fully from the bottom up."
Power is an emergent social fact.
It’s ground truthed, ultimately, in “who could vanquish who (and survive!) if it ever came to that.”
But of course it almost never comes to that in modern society.
That means almost all power games are proxy battles.
Part of dominance is acting dominant and having others believe you.
Dominance displays that are convincing often are hard to fake.
The Handicapping Principle.
For example:
How strong are you?
How nimble and dextrous are you?
How much of a posse will show up to back you up?
An insightful HackerNews comment about how hollowed out modern society is:
"The problem isn't retirement per se, it is that people don't have things to occupy themselves with. They retire and they vegetate. I worked with a lady that was in her 70s who was deathly afraid of retiring because she didn't have anything to do. That's beyond depressing to me, to be incapable of even conceiving of doing something that doesn't involve going to a job.
We have created people that never develop as human beings outside the context of their being economic entities in the workforce and that's not something to celebrate."
Whenever you care about the means more than the end, you’ll be incentivized to cheat.
Belief in some collective meaning is the only antidote to cheating.
A society with low trust will corrupt itself, with the only way to get ahead to cheat.
"Most Americans Would Give Up Nearly Their Entire Paycheck For Peace Of Mind"
The Cacophonous Era.
Fascists are obsessed with order.
Specifically, their own concept of order.
All of us are a little obsessed with the kind of order we care about; a fascist takes it to the extreme where the order is more important than anything else.
Fascists don’t believe in beautiful emergence, only malignant emergence.
All emergence is malignant to them, because it goes against their conception of what should exist, and that is all that matters.
People like working on things that converge.
Cleaning up, for example after a party, naturally converges.
Everyone can agree what clean looks like.
Setting up, for example for a party, you don't know what the organizer wants to have happen, so it doesn’t naturally converge.
When you get married the relationship goes from default diverging to default converging.
Default diverging before: the relationship could end in a moment unless you keep it going.
Default converging after: the relationship will continue going unless you take significant action to stop it.
Curiosity comes down to: does ambiguity feed you?
Herb Stein: “If something cannot go on forever, it will stop.”
Obviously true, but easy to forget.
An idea I heard about Costco: increasing prices by just a little is a hit of heroin.
They could raise prices by 3% and double their income and no customers would notice or care.
But if you do it once, you’d do it again.
And a few times later, you’ve destroyed everything you stood for.
You’d torch your trust, incrementally.
The trustworthiness you’ve accumulated is your most valuable asset–and others who don’t care about your mission will see it as pile of gold to plunder; a stack of kindling to set on fire.
When you maintain your principles even in small ways, it generates trustworthiness.
People can see you sacrifice for the principle even when it’s hard, making it more likely you’ll do it in the big ways, too.
That creates stores of trustworthiness, and that loyalty is powerful.
Every time you run into a challenge with the principle, it gives you an opportunity to show off your principle you're proud of.
One of the reasons the $1.50 Costco hotdog is so powerful.
A study in 2020 found the economic costs of short-termism: 5.8% of market capitalization.
The headwind was not due to compliance cost, but from cultural corruption that came from optimizing for quarterly returns.
The quarterly reporting sets the horizon of how long-term teams can easily think in an organization.
The SEC is currently considering a proposal to allow companies to opt-in to reporting only every 6 months.
It was the Long Term Stock Exchange that lobbied hard for this regulatory change.
All stock exchange listing criteria are basically the same.
Why isn't there competition?
Those set the physics for what kinds of companies are possible.
If your mission and business model are not aligned, you'll tear yourself apart.
A PBC charter puts your mission into your charter, so they can't be at odds.
An observation: "Making money without creating value is corruption."
Optimizing for profit allows corruption to achieve it.
Mistaking the means for an end.
Profit is the proxy for value creation for society, but it is only a proxy.
Do you live by your principles even when no one is watching?
Don’t be an OK version of what someone else thinks you should be.
Be a great version of what you think you should be.
When you’re grinding on your own path to greatness it’s fun.