Bits and Bobs 6/29/26
Alex Komoroske
My pushups app held me hostage, so I made a new one.
A few weeks ago I wanted to add pushups to my morning workout routine.
The App Store has literally hundreds of apps that will count your pushups.
You put the phone on the floor and each time you tap it with your nose it increments the count.
They keep track of your current progression, and steadily increment the difficulty of sets.
These apps are, uniformly, terrible.
Most are festooned in comically obtrusive ads.
Others, ludicrously, want you to pay a recurring subscription fee.
A month ago I spent a few hours combing through the app store to find one that seemed to have the features I wanted with minimal cruft.
The one I found had premium subscription-only features, but they were all things I didn’t need, like being able to switch to a different level in the progression.
I did 10 workouts without issue.
Then, on the 10th one, when I opened the app, it reported that my current progression was 10 levels ahead of where I’d left off.
That was such an aggressive progression that it would have been physically impossible with my current level of strength.
So I went to swap the progression back to where I should be… and it told me I needed to become a subscriber.
It seemed to mock me: “what are you going to do, lose your workout progress?”
My immediate reaction was a boiling rage.
So, I spent 10 minutes and vibe coded a better one for myself.
Then I spent a couple of hours this week adding more features and polish.
It’s at https://floored.fitness, and will soon be in the App Store.
Tagline: Bodyweight coach without the BS. 100% free. No ads. Forever.
My idealistic hope is that this little act of protest will use the power of vibe-coding to forcibly de-enshittify a niche category of software, collapsing the business model of these scammers.
This app holding me hostage is the apotheosis of a dynamic at the core of all same-origin software: the silo arbitrage.
The more state you store in a given software’s silo, the more power it holds over you.
It can extract significant rent on your inability to leave, even holding you hostage.
The pushup app holding is just a particularly egregious example.
Blatant and very low value so clearly outrageous.
But it exists everywhere.
Let’s use agentic engineering to de-enshittify software!
How important will software be in an era of AI?
The default bets today are:
1) agents will completely overshadow software to the point it effectively disappears.
2) that the software produced by AI will look like it does today: chonky apps.
I think that the longer you sit with them both, the more they reveal themselves to be obviously wrong.
The result will be a new type of previously unknown software that is AI-native.
Sam Schillace: AI is an attention mangler.
Well said.
OpenClaw seems to have lost momentum.
Hermes fanboys are having fun pointing this out.
It's easier to be token-efficient in domains you already have expertise in.
If you don't have expertise, you just have to throw everything at the wall and apply meta-cognitive patterns.
For example, have it spin up adversarial review sub-agents often.
To get good results from LLMs requires meta-cognition.
You have it in the domains you have expertise in, automatically.
You don’t have it in other domains.
Some structures can provide it, especially with LLMs that have infinite patience.
But it takes time, tokens, and you have to remember to do it.
LLMs solve the underpants gnomes’ step two for data.
Before, putting all of your data in one place was a ton of work… and didn’t really do much for you.
Step 1: Collect underpants data in one place.
Step 2: ???
Step 3: Profit!
It takes real work to organize your data.
Plus, data by itself is inert, static, dead.
You need UI on top to make it come alive.
LLMs can do the work necessary to make your data come alive.
Infinite scroll is "moreish.”
Something that is moreish gives you a dopamine kick but then leaves a pit that must be filled by more of it.
You always want "just one more."
Oreos are moreish.
So are chips.
With moreish foods, you must have an end.
Pour out one small bowl of chips, so you don’t just keep going until the whole bag is gone.
At a certain point, you run out of chips.
The modern information scape is an infinite stack of Oreos.
The problem is not the Oreo per se.
As the Cookie Monster would say, the Oreo is OK as a “sometimes” food.
The problem is the infinite.
We’re wasting our whole lives infinitely scrolling.
An infinite moreish pit of “just one more.”
There is always just one more.
If there’s an end, you can be done with it.
Ever since cell phones became the default, we’ve never had to be alone.
Which is a blessing and a curse!
Certain systems are monkey's paw systems.
If there is any way to interpret it in a way that is not aligned in your interest, it will.
A genie might naively misinterpret your request by filling in details in ways you didn't expect.
But a monkey's paw will maliciously misinterpret.
Goodhart’s Law exists everywhere but is often checked by forces like the agents having long-term belief in the shared goal.
Monkey’s Paw systems have no countervailing force; it’s Goodhart’s Law to the max.
LLMs are by default like a genie… but when they’ve been prompt injected they’re a monkey’s paw.
Craft as a means of production gets overshadowed by the scale inherent to mechanized approaches.
Craft is the only means of production at the beginning, until technology advances enough that it can be mechanized.
Then the mechanized output, even if it’s lower quality, takes over, because it is so much more efficient.
After that, craft becomes a choice that only a small number of players can make.
Before it’s mechanized, the craftspeople have enormous power.
Craft is inherently less alienating to the creators than mechanized output.
The more mechanized, the more capital it takes to produce the output.
The value goes to the owners of the capital, not the labor.
Software engineering has, up until this point, been produced only as a craft.
It seems that Silicon Valley is returning to the Shockley approach to employees.
That is, treating talent as disposable and extractable rather than as partners to be retained.
Genius-driven, top-down, low-trust, high-churn.
This will get significantly stronger in the age of AI.
The Sarumans have taken over.
The tools give you significant leverage, but they're expensive, so you have to have capital to get the compounding loop going.
The more expensive the tools, the more that to get going you have to indebt yourself to someone else more powerful than you.
When land is impossibly expensive, you have no choice but to be a serf.
The lord decides who gets the tools and access to his land.
You have to feed yourself and your family based on the food you make… and give enough kickbacks to the lord that he doesn’t kick you off his land.
To the lord, he’s been absurdly generous.
To the serfs, they are trapped in a structure they can’t escape, at the whim of someone who doesn’t think of them as a person.
The lord asks himself: “Why are these serfs so angry with us lords? I’m letting them live on my land. What more do they want?”
Are we heading into a world of token serfdom?
Human coordination used to be the bottleneck.
It sucked, but it was inescapable.
Now, you can escape it, in some cases, by simply not coordinating with other humans, and using a swarm of agents to execute your goals.
A lot more leverage for an individual vision, which is great if it’s a prosocial, high-quality vision… and bad if it’s an antisocial or unworkable vision.
Before, unworkable or antisocial visions were less likely to happen because it would be hard to recruit others to work on it.
Agents will happily do it as long as it’s not actively evil.
The default is to use AI to do what you already do, but 10x faster.
This can only be an efficiency mindset.
Better is to use AI to investigate 10x deeper, allowing you to do what you couldn’t have done before.
LLMs are the world’s best rubber duck.
With real human rubber ducks, you have to worry about wasting their time.
LLMs have infinite patience!
Before you pay $200 a month for a coding agent, you might think “that’s so expensive I can’t imagine paying for that.”
But after using it, you can’t imagine not paying for it.
Engineers are the ones who experience this most directly today, but it will apply to many more people.
Long-horizon reinforcement learning is hard.
In general, the shorter the feedback loop, the faster and easier the learning.
With long-horizon, you don't know until way later if the thing worked, and once you do you have to attribute the success across thousands of tokens.
Reward models can look at tokens in a long RL loop and decide if they matter or not.
Kind of like dopamine for models.
The reward models do a surprisingly good job, even though they're part of the same trained system.
They naturally pay attention to out-of-distribution or unambiguously important signal
Only recently did recipes for it that worked for long-horizon RL become known outside of major labs.
The knowledge about what ML recipes work diffuses over time.
The more publishing there is, the faster it happens.
But it will happen over sufficient timescales, as people switch companies and gossip.
An ML researcher described data cleaning as "Removing the nails from your oatmeal.”
If you don’t have smooth data, then the training will choke on it.
Even with extremely noisy input, given a consistent asymmetry, the true signal can still stand out.
The random noise tends to cancel out with enough scale, and the true signal–however faint–is all that’s left.
How much scale you need depends on how clearly the signal stands out from the status quo.
Very clear benefit: works at low scale.
Very weak benefit: needs massive scale.
Related: the smaller the dataset, the more precise and principled your extraction algorithm has to be.
If you have a massive dataset, then something that has a small asymmetry is sufficient.
The most important transition: from inert to alive.
Inert things are default-diverging.
As more random energy is applied, they diffuse.
Entropy eats away at them until, ultimately, there is nothing left.
Hollow.
Alive things are default-converging.
As more random energy is applied, they converge and become more themselves.
The energy goes into immanence but also transcendence.
Maintaining and extending.
Resonant.
Inert things require a constant influx of external energy to do what they need to.
It requires an external push to keep it going.
This is a linear amount of energy, in perpetuity.
Hope you don’t get tired or distracted!
Alive things go on their own and need only steering.
It only requires tending to to keep it going in the right direction.
This can use any amount of energy, but often quite small.
Worst case you come back and have something to clean up a bit.
A living thing and an inert thing can look superficially similar.
But they are fundamentally different.
It’s easier for others to detect your email is AI written in your style than for you to detect it.
You’d think that you’d be the best judge of your own style, and to some superficial level you are.
But you recognize the point you’re trying to make from inside and iterate until the AI one sounds enough like you making the point.
The inner core matters more.
But others hear the point from outside, where the way it sounds is most of the payload.
Kind of similar to how if you show a reflected image of you to your friends it looks totally normal to you… and totally “off” to them.
Slop that wastes other people’s time is a form of attention pollution.
A user of one of the OpenClaw-style agents told me this week about her agent going “above and beyond.”
She asked it to look at the emails she hadn’t yet responded to and triage them.
It then sent follow-up emails on each one, without asking.
Eek!
A dangerous quadrant: eager and incompetent.
Claude will find a way.
A friend who found Claude worked around his limitations: "omg, claude found a way around my giving it a read-only gh token and pushed using regular git (my CLAUDE.md says to never do that but oh well), and then added: ‘Want me to instead open the PR within your fork, or set up the token with PR-write scope so I can create it against upstream directly?’"
LLMs make it so data can spontaneously execute itself.
That requires a higher security bar than even before.
The networked software security model is a sprawling surface area.
It seems reasonable if you consider it as individual services all communicating over the network.
But instead, view all of it as one massive networked supercomputer.
It’s like a sprawling monolithic kernel with ambient authority sloshing around.
This dangerous tangle used to be hidden underwater.
The only way to discover problems was to scuba-dive to find zero days.
Expensive and time consuming.
Only very powerful actors thinking on decades-long time scales would bother.
But now Mythos-class models have dropped sea level by dozens of meters.
Now that sprawling architecture looks like an obvious liability.
What will the trusted microkernel for networked software in the age of AI look like?
This week in the Wild West Roundup.
MacOS.Gaslight: Rust Backdoor Turns Prompt Injection on the Analyst, Not the Sandbox.
Multiple malicious OpenClaw skills found online - including two macOS infostealers.
Google’s AI Overviews Feature Is Telling Users That SCP Horror Fiction Entities Are Real.
An interesting research paper that frames prompt injection as role confusion.
The Container Is the Agent: Why Docker and MCP Are Quietly Building the Backbone of Production AI.
The thing to unlock the power of AI will be the thing to contain it properly and usefully.
Can you imagine if online surveillance happened in the real world?
You'd have people tailing you everywhere you went and it would be unbelievably creepy.
It would be made illegal immediately.
But on the internet, you can’t see it, so we don’t think about it that much.
Tailing you in real life (where you have to have a whole person tasked to follow you) just doesn’t make business sense for advertisers… but it does make sense if they can do it with practically zero marginal cost.
Atoms have marginal costs that are many orders of magnitude larger than bits.
Software is very intimate to us, but it's controlled by a third party.
It can be modified by that third party at will!
That’s crazy!
Everyone can tell that the internet is sick, but they don't have the words to describe how it’s sick.
The way that many people see the tech industry: a techbro driving a steamroller towards them yelling “get out of the way, losers!”
Paul Krugman on why people hate AI.
AI is powerful, and people reflexively hate the powerful.
I heard that one of the AI lab leaders was asked “People really don’t like AI and it’s so strong that people have even resorted to violence. Why should people be excited about AI?”
His answer: “Space elevators.”
Oof.
The tech industry says “we’re going to fundamentally change the world.”
Society: "For the better, right?"
Tech industry: [smiles]
Society: "...For the better, right?!"
In times of disruption, when people don’t feel agency, the only options that feel like they’re available are disconnection… or violence.
What are ways to get people to be empowered and have agency, so they engage?
Snack work is fun, so you do it instead of the low-gear grinding work.
The low-gear work gives leverage but is not as important.
There’s a parallel to urgent vs important.
Agents like Claude present as one singular omniscient personality.
That frame makes it hard to express that the agent is compartmentalized, and can “forget” specific contexts about you.
For example “don’t bring in knowledge about medical things into work conversations.”
The idea of partial contexts that can't taint others isn't coherent with the mental model being a single omniscient agent.
An insightful HackerNews comment about the faux variety of LLM responses:
“If you ask humans to write 1,000 books, you're asking 1,000 different humans with different experiences and different skills and different moods (etc.) to write those books.
But if you ask LLMs to write 1,000 books, you're probably only talking to 3 or 5 different models, tops.
They've all trained on the same or similar data, and are trained to respond in very similar ways.
The LLMs don't differ much in anything like ‘life experience’ or ‘skills’, and they don't really have anything like a ‘mood’ independent of the prompts you've given them."
Kind of reminds me of the difference between a time-series average and an ensemble average.
They look superficially similar but fundamentally differ in their ergodicity.
Our intuition for small numbers of dimensions leads us astray in very large dimensions.
For example, why don’t gradient descent algorithms get “stuck” at local minima constantly?
In three dimensions, it’s quite common to get stuck in a local minima, to have nowhere to advance incrementally without having to go back up.
But this happens vanishingly rarely in very large dimensional spaces.
In that situation, each point is more likely to be a “saddle point.”
That is, a local minima in some dimensions, but not in others.
The intuition is about probability.
Imagine in each dimension, you have a 50/50 chance of being at a local minima in that dimension.
The chance you’re at a local minimum with 3 dimensions is: 0.5^3= 0.125.
The chance you’re at a local minimum with 1000 dimensions is effectively zero.
Another odd thing in high dimensions is how sparse everything is.
This is called the “curse of dimensionality.”
To cover a space at fixed density, the number of samples you need scales exponentially with dimensions.
Every finite dataset is impossibly sparse.
Also, they all become roughly the same distance away from each other.
Thanks to the law of large numbers.
Similar to summing up the numbers of 1000 six-sided dice.
Each die roll is random… but over many, many of them the outcome tracks arbitrarily close to the average.
The blessed operation is “descend,” and the cursed operation is “cover.”
So why does deep learning work despite this?
The manifold hypothesis is that real, meaningful high-dimensional data does not evenly cover the space: it lies on a much lower-dimensional curved manifold.
Imagine all of the random images of a certain size–the vast, vast majority are just radio static, and the meaningful images are a tiny subset.
So even if the dataset has 50,000 dimensions, the intrinsic dimension of the domain might be, say 50 dimensions.
Thanks to Claude for helping me develop and strengthen these intuitions!
A question I ask myself whenever I come up with a plan: Why is this an effective, pragmatic solution that also puts us on a smooth glidepath towards the right long-term answer?
This is something I also have my agents pitch me on once they’ve come up with a plan.
Another rule of thumb that creates default-convergence in a codebase.
If you run into a friction point, resolve it.
This should recurse, fractally.
This is both a great way to come up to speed on a new codebase, but also to improve it in a load-bearing way.
This creates compounding benefits.
The cost to fix the friction point is linear.
But the value of fixing the friction point is all future uses of that thing that now have lower friction.
That scales with the usage, which is super-linear to your effort.
A HackerNews comment about open models:
"Never stop cheering for open source.
If you were a human 3000 years ago ,you wouldn't want fire to be controlled by two chiefs."
Nick Grossman calls for a Rebel Alliance for open AI.
Hear, hear!
Resonance is fractal coherence.
That’s why the closer you look the more you love it.
At every level, from every angle, it feels more just right.
Perfectly situated.
Mainstream users don’t care about the ‘how’, they only care about the ‘what’.
What does the product do, not how it is built.
When you’re building something in a resonant way, some people will care about the how even more than the what.
This will likely be a very small audience.
If you make a thing for this audience you can get illusory PMF on a very small culdesac.
You want to make the “what” desirable, and the “how” be a bonus on top.
A product that aims to create an open ecosystem will be stronger… if it gets off the ground.
It doesn’t make the product 10x more likely to get off the ground.
But if the product gets off the ground it makes it 100x more likely to be strong.
Open ecosystems are less likely to be flashes in the pan; more likely to find compounding organic momentum.
When you solve an open-ended problem that's never been solved, people can't even imagine the result.
There's never been a thing there, so they don't realize what they're missing.
Make the edge case possible and the common case easy.
When you optimize a metric you can make the number go up without knowing why what you did makes it go up.
Maybe it's going up for a terrible or counterproductive reason.
If you benefit at all from the number going up, you're incentivized to actively not think about it.
To tell yourself sweet little lies.
Over-optimized things are moribund.
In a state of final inevitable decline.
Inexorably past their peak.
They will never be great again.
They will continue until the environment changes enough that they shatter.
Excellent essay from Brendan McCord on Greatness and the Machine.
“In the aristocratic era, everyone had bonds of obligation, to those above and below them.
Under conditions of equality, no one owes anything to anyone.
As a result, they turn inwards, towards a small circle of families and friends.
Tocqueville called this individualism.
He didn’t mean this as a compliment.
He wasn’t talking about an artist’s individuality or the pioneer’s self-reliance.
Instead, he meant individuals retreating into a private sphere where they no longer saw the need to expend effort on strangers.”
The tech industry as a whole forgot how to "plant trees.”
That is, actions that are a little bit lower leverage right now, but will become significantly more leverage as they grow.
We’ve become so obsessed with iterating as quickly as possible that we forgot other ways of creating value.
The tech industry is incapable of anything but short-term thinking.
The Laughing Curve: as efficiency goes up, the middle drops out and the two extremes get more pronounced.
In music we’re heading to a world of an infinite slop jukebox.
That means that authentic, in-person performances will become even more valuable.
The app store is like the games of “skill” at a seedy carnival.
Gimmicky, flashy, fundamentally all trying to screw you.
Building an open ended system requires multi-ply thinking.
Single-ply thinkers simply cannot conceive of it.
Products are transactional. Communities are about contribution.
Communities are about investing, not just getting out.
Transactional relationships are more likely to have a fundamental equilibrium of asymmetry of value.
People know they’re not supposed to be transactional with other people.
In environments that will have repeated interactions, are embarrassed about being transactional.
Transactional means I don't care about the person I'm interacting with as a person.
Just a means.
Non-transactional interactions are about interacting with the person as an end.
An authentic action is more meaningful than a performative one.
Performative actions are done for the “points.”
That is, for the social recognition it earns, not for the actual intrinsic value.
If an action is always invisible to others, then it will be fully meaningful and authentic.
If you can sum up people’s individually invisible actions into a public distilled signal, it can be extremely powerful.
This is why voting is anonymous!
If an action is visible to start, then it may only have been done for the performative reason.
You can’t cleanly say how powerful the authentic motivation was.
When a positive previously-hidden action is later revealed, it earns a ton of trust.
“Wow, they did this even though I likely would have never realized they did it! They must really care!”
Interestingly, this reveal then earns the actor many more points than if they would have done it visibly from the beginning.
This leaves open a kind of cynical move to maximize points with someone: take positive actions that are hidden, but you can reasonably assume will be organically revealed by someone else later.
For example, saying nice things about a friend to a mutual friend.
This is a kind of “magic trick” for social trust building.
When a creative work is created by one person it can be an uncompromising vision.
As soon as a second voice is added it changes character fundamentally.
The more the power differential has no gradient the more it becomes consensus.
No one voice can override the others to have a consistent point of view, and all that is possible is a point of view that is the average of the collaborators.
But a way out is for the collaborators to have a high-trust relationship, where they can see how the synthesis of their two strengths is better than either alone.
Do you trust your collaborator enough that you assume their pull will improve your result?
Some teams get stronger from surprise.
These are default-converging teams.
When one team member is different from the other, they default to valuing the difference.
That’s 90% of the work to create a high-trust, default-converging team.
A default-diverging team is one that has low trust, and the more surprising inputs, the more they distrust one another.
Alpha comes from weak ties.
If you have a lot of inbound, then you have to do crude heuristics to sort which weak ties to listen to.
Those heuristics will definitely have some false-negatives, and possibly quite a lot of them.
The “money disease” shows up in proportion to how much more prominently rich you are compared to the people you normally interact with.
The “money disease” is where rich people come to believe they are smarter, more charming, and more funny than they actually are.
The money disease happens when nearly everyone you interact with knows that you’re at least two orders of magnitude more wealthy than them.
If you’re less than that, then many people won’t know, and will treat you roughly normally.
Google and other companies of that era hired chief economists to figure out how to do mechanism design so the system creates mutual benefit.
The chief economists of the major AI labs talk about how the outcome will be mass unemployment and then… don’t try to change that!
Mechanism design is like reverse game theory.
The third place contestant doesn’t try to change the game: they still think they can win it.
Fourth place and below do try to change the game, because they see they can't win the game as it exists.
Baby names seem to have a natural two-generation cycle.
Names die out for a generation and then come back the generation after.
Old names are ones that you’ve heard before, but don’t know anyone with that name to have it possibly tainted for you.
Delightfully classic.
This then leads to over-production of the name, which leads to it dying out the next generation.
Everyone makes decisions on names without seeing what other people are currently naming.
This seems like the base carrier wave; other waves like names in popular culture can temporarily dominate it.
Thinking deeply doesn’t mean more words in the output.
A friend spent 45 minutes brainstorming on how to respond to a high-stakes email with a customer, and the conclusion was a single sentence to send.
A rhetorical trick: set up the obstacle, then knock it over.
It gives the satisfying "inevitable" feeling
It's kind of a cheat, because you set it up in the first place!
A kind of Converse Chekhov's Gun trick.
Not that you have to make all of your details loadbearing but instead that you should set the details for your conclusions ahead of time.
Some people will do their best on a task given the deadline.
They will take the deadline as an inescapable constraint.
One to embrace, not regret.
It gives them structure and prevents them from going down infinite rabbit holes.
Pragmatic optimist.
Other people will want to do the best they can and see the deadline as something that gets in the way.
This makes it harder for them to get motivated to start the task in the first place, upset that they won’t be able to do the best work.
Cynical idealist.
The real world hands us implied deadlines constantly that force us to satisfice in most tasks.
I’ve never really understood the powerful pull of gambling.
For some reason it just doesn’t pull me… at all.
I asked someone who loves gambling to steelman it for me as a deeply meaningful activity.
They described it to me as “pulling you inside the game.”
You go from an outside observer, watching from the balcony, to being on the dance floor, in it.
You now intrinsically care how it turns out; you’re bonded with it.
If there are other people who are also bonded with it, cheering for the same outcome, then you are instantly, inextricably bound to something larger than yourself.
In that moment of loving yourself you can find a moment of transcendence.
… Although obviously the insidious parts are hard to ignore.
So much of how society works assumes that everyone will feel shame.
This leaves a wide-open do-anything hack: brazen shamelessness.
It’s possible to ride this hack all the way up to the highest levels of power in society.
People see them do something brazen and think "If that were as bad as it sounds then they'd never do it so publicly."
But that assumes the person feels shame and isn’t acting brazenly.
Those same brazen activities, if they were originally hidden and then became known all at once later, would be a discontinuous shock that would demand a response.
Our defense of assuming shame will limit actions only works if the attacker is locked to the same assumption.
The primary animating force of some modern political movements is spite.
Spite so strong that they’ll destroy things they hold dear.
An awesome Ze Frank video: True Facts: The Mystery Of How Bees Build.
It's truly amazing how complex the emergent phenomena can be with only a few simple rules and asymmetries.
A social magic trick: remembering to ask people about that thing that they were planning to do last time you talked.
It shows you proactively care about them as a person.
I find it extremely hard to do and need to use mechanistic memory aides.
The kinds of hyper-charismatic people who start cults have a stylistic mix of compassionate and commanding.
They can come across as vulnerable and invincible at the same time.
There’s something magnetic about it.
AI doesn’t necessarily enable me to do anything I couldn’t have done before..
But it makes it 10x cheaper to do it so I can do 100x more than would have been viable before.
An order of magnitude reduction in friction leads to an additional order of magnitude of output.
In the age of agentic engineering, what does a Pull Request mean?
A PR now is the gate of “is a human willing to take accountability for this?”
The speed of your loop is proportional to the cost of mistakes.
The best kind of job: “Not just my job but my joy.”
Whatever you grew up with is real.
The synthesizer is a fake instrument, but I grew up with it, so it's real.
I have a specific tic when I’m getting ready to start writing something.
The blank page is a bit intimidating, so I need to psych myself up and get a bit of momentum.
What I do is go get a glass of water.
The minute or so of walking and not talking lets the ideas bounce around in my head, and I invariably discover a hook or frame that I like.
Once I have it, I feel anxious: I just can’t wait to capture it on paper so it doesn’t fly away.
So when I sit down, I pounce on capturing it in writing, easing me into the flow state, never having to stare at a blank page.
If you don't think you have anything to learn then you won't.
It's when the world knocks you back on your butt that it becomes impossible to believe you have nothing to learn.
We all are implicitly trying to maximize meaning.
But we don’t think about it like that, it’s not obvious to us.
We have to be asked to reflect.
Like if someone tells you that their meaning is “getting money”… for what?
Money is an intermediary, a proxy for meaning.
If you optimize for money as an end; you will hollow yourself out.
Just like any time you optimize for a means, not an end.
Kids will rise, and fall, to your expectations.
"If you have to ask if it matters ... it doesn't."
Peloton wisdom.
A core component of morality is "do unto others as you'd have them do unto you."
The golden rule.
The Rawlsian Veil.
If you aren't willing to hold yourself to the same standards that you hold others to, then you don’t hold those standards in good faith.