Bits and Bobs 3/25/24

The last week I spent primarily on saying goodbye from my last role. If you're coming from the LinkedIn post and looking for the AI-adjacent stuff and hints for what I'm up to, check out last week's snippets first.

What's the biggest thing I've noticed so far?

The quiet that comes from being disconnected from Slack.

Slack is a background cacophony, a constant reminder that there are things happening all around you in your organization.

At its best, it's the kind of hustle and bustle background sound of a thriving city.

A kind of pleasant sound of things happening.

But more often, it's just the mosquito buzz of the urgent, never leaving your ear, making you anxious at all times, telling you to stop thinking about the important, and to focus only on the urgent.

In large organizations, this can become a background roar.

The other thing I've noticed: for all of the crap LinkedIn gets as a "social network", the professional positivity when you announce a big life change really does feel great.

I feel like I'm at the beginning of my hero journey (before we know if I'll be a hero or die trying), setting off into the intimidating forest.

Everyone I've ever interacted with back in the village is cheering me on, buoying my spirits as I tackle something terrifying and novel.

Seeing all of the faces of people I've interacted with over the years and knowing they're cheering me on… that feels great.

The things that make something customizable make it hard to handle in a scaled / levered fashion.

Centralization creates so much efficiency that as a user or maker you have to exert more effort to stand out from the gravity well of the cheap way everyone else is doing it.

Since centralization wins against customizability in each little micro paper cut decision, it keeps on compounding.

Before you know it, you get one-size-fits-all software for everyone all the time.

This happens even if people still want customizability.

The logic of aggregation simply gets so much momentum that it steamrolls everything else.

A net loss of value for the ecosystem.

Where's the hyper-customizable software for power users?

What was the killer use case of the web? Google.

Google wasn't even imaginable until the web was already deeply underway.

It was a thing that had to do with the specific novel dimension of the web, the thing that was most unlike other systems.

The web was different from software from extremely low friction and ability to visit novel, untrusted content immediately.

So the killer use case was a search engine to make sense of that diverse, cacophonous ecosystem.

Each individual website is less important than the connection between everything into one whole.

Websites are stronger together.

As soon as a page has a link in (or out) to another part of the web, it's tied into something much larger than itself.

This means that much smaller experiences are viable (e.g. pages about your random hyper specific hobby).

That means that there's more room for experimentation, and collective innovation.

The web is a collective, emergent project by society, not owned by any particular entity.

That implies that we're still in the community garden phase, not the factory farming phase we've been used to for the last decade of pre-LLM technology.

This is also one of the reasons that despite the clear value of LLMs, some people who have tackled it from the perspective of companies are erroneously concluding they aren't valuable.

At the beginning the skillset has to be camera-making, and over time it shifts to creative skills as you can take cameras for granted.

When cameras came on the scene, first portraiture went away in paintings, but later paintings became more impressionistic and also abstract, because they were freed from the job of realistic depiction.

Just like Google-fu back in the day.

The kinds of people who had developed an intuition on how to formulate their query (sometimes in non-obvious ways!) to get great results.

In the early days of Google, some people literally got paid for this skill!

The people who find LLMs most useful today are meta-experts: generalists.

People who play "out of position" a lot.

Asking a domain expert if an LLM gives a useful answer in their domain of expertise is the wrong question.

The question is: for the generalist, does it give them more calibrated confidence in an unfamiliar domain more quickly than they would have otherwise?

At that point it's a race to who can attract enough capital, execute most frenetically, and gain an edge in any way possible.

For example, after Bird scooters 4 years ago showed there might be a there, there, there were a dozen or so competitors within months.

How much this dynamic happens boils down to: "given the elevator pitch, how well could the median entrepreneur execute on it?"

If the answer is "quite well" you'll have a tiny amount of time to get a head start and then it will become a ruthless red ocean.

Execute in stealth, try to get the biggest possible lead before everyone else notices.

If the answer is "not at all," then once you clear the "there might be a there, there" threshold, it's blue ocean.

It doesn't matter who you tell, because no one else can execute the idea anyway.

It requires a subtle, nuanced cocktail of specific knowhow.

The more people you tell, the stronger the idea gets, as the idea builds in people's minds, as you get disconfirming intel to grow it stronger.

There is a huge asymmetry of a big idea that only you know how to execute.

The most differentiated ideas are ones that everyone can grok, on some abstract level, quickly... but no one else can execute.

The fuzzy, amorphous, ever-shifting gray goo of the real world is hard to "grab onto."

A name gives a label, a handle to a related set of adjacent concepts, and says "these adjacent globs are a Thing".

The name starts out not meaning anything, just a convenient handle.

But as the name is used more and more broadly, people's expectations of what comes with the name start to matter more.

People watch how other people use the name, and they update their expectation to be closer to that, too.

The name includes, implicitly, where the boundaries are between the Thing and Other Things.

Over time, the name becomes a kind of rut, hard to change.

Those emergent expectations outside of the Thing can box in the Thing.

Once the Thing becomes really well known, it might start to stand in for a Category of things, and really become a box.

"What kind of thing is it?" doesn't have a satisfying answer.

People are used to being able to say things like:

"Uber for dog walking"

"Vertical saas for funeral parlors"

That is, "<pattern> for <population>"

When you can't do that people say "I guess the idea is very abstract."

But the idea can be extremely concrete, just big and new enough to be hard to grab onto.

The only way to grab onto it is with metaphors, but those can be confusing.

"It's a metaphorical equivalent of the browser for a new kind of web-like fabric of experiences."

"Oh, OK, I get it. … But wow, browsers are expensive to build and maintain!"

"Well, it's not literally a browser… it will manifest to start primarily as a special kind of web app"

"Oh, OK, I get it, so it's a web app."

"No, it's a special kind of meta web app…"

… And so it goes until the conversation dies out.

Truly new things can't be talked about; we don't have the labels and pre-existing expectations yet.

Truly new things simply must be experienced to get it.

The pitches that are hardest to distill have the most asymmetric upside.

… That is, assuming they're good ideas. They could also just be incoherent noise.

Novel ideas and noise are equally challenging to compress.

The willingness of people to sit with the noisy idea and try to make sense of it has to do with how noisy it is and how much credibility the person proposing the idea has.

The weirder the thing, and the less respected the author, the less willing people will be to absorb the idea.

An example from holding your breath.

The way our bodies decide how badly we should take a breath is the density of CO2 in our blood.

But the actual thing that determines if we need to breathe is the inverse density of O2 in our lungs.

In the vast majority of cases, these two signals correlate very well.

But it's possible for them to diverge!

For example, if you hyperventilate before going underwater and holding your breath (to get more air in your lungs) you expel CO2 better than you take in O2.

This can lead to a situation where your body doesn't realize you need to breathe and you asphyxiate underwater.

Another, less morbid example: "I don't have an easy way to grab onto this idea, therefore it's abstract"

Novel things means, "something unexpected"

If a thing is omnipresent and never changes, then it literally fades from our awareness, we don't pay any attention to it.

We simply take it for granted.

In a lot of cases, this makes sense, because the fact it never changes means we don't have to consider it changing.

But this is distinct from the force being important.

Gravity is an extremely important constraint, but it never changes in our day-to-day lives so we simply don't think about it.

But this sets us up for nasty surprises if it does change.

For example, imagine it breaks at some moment; suddenly all of your baseline expectations shatter for what is possible, because they all assume this unmoving baseline.

Like assuming high-bandwidth internet, and then having an unexpected outage.

But even worse is the subtle breakage.

Imagine being locked in a windowless room for your entire life.

You develop intuitions about how things behave.

But then the room tilts 2 degrees on its axis.

Everything looks the same, but is now subtly shifted.

It will feel a bit "weird" but in a way you can't name.

And now your expectations will be consistently wrong.

"I don't need to worry about that stack of fragile objects, it's never fallen over before"

… and then you put another thing on it and the higher tilt leads it to fall.

When this happens, your priors are all wrong, and your intuition is off.

This also happens when a technical paradigm shift is underway.

"Of course this is right, we've always done it this way!" works great until the world tilts on its axis.

Agency implies that it can do something behind your back.

Any time you execute code you don't deeply trust, it has agency.

It can stab you in the back.

When code runs on someone else's turf, you can't deeply trust it; it has its own agency.

If you're in a scarcity environment, it's much harder to get out over your skis; the feedback loop is tighter.

Googlers typically don't understand how to do anything in a scrappy, scarcity mindset given how resource-rich the environment is.

The playbook I used to be successful within Google, asymmetrically, was to adopt a scarcity mindset and do lateral thinking with weathered technology.

Also written up as The Live Oak Playbook.

That playbook in an environment of plenty is an asymmetric advantage because it leads to such different tactics and goals.

That playbook is resilient; it works well in many different contexts.

The project is invested in to an almost absurd degree, with absurdly large goals to reinvent everything, and then collapses under its own weight.

But the knowhow and technologies developed in that project can still go on and be useful inspiration for more realistic systems later.

For example:

Microsoft's Cairo in the 90's, which gave COM, Plug-and-Play, and others.

Google Wave, which gave Operational Transform.

Google Turquoise

These kinds of high-profile failures are like a whale fall, a bonanza for scavengers.

Lots of new components to do lateral thinking with weathered technology with.

Obviously as the org scales you get the dreaded coordination headwind.

But it also encourages you to try to boil the ocean; to complicate too many things and try to power through too many walls.

The overall project grows in complication and smothers itself under its own weight.

A powerful meta strategy: take big, ambitious visions from previous failed mega-projects… and approach them in a "lateral thinking with weathered technology" mindset.

Same vision, different path there.

One that requires it to be grown, not built.

Tending to acorns, not building massive edifices.

You want a simple scrappy thing that is alive, not a large, over-built thing that is dead

If you find a cheap thing that is alive and can grow into the massive vision incrementally, that is highly asymmetric.

While growing it has to be viable at each step.

Or in evolution, each intermediate adaptation also has to be useful enough to be maintained.

A path from there to there that is continuously viable.

Built things don't necessarily need that.

But a built thing will erode; it needs living things to decide to invest in, who find it useful.

And if you built a thing, you don't know if you'll find enough other things that want to invest in it.

Whereas a living thing, that has been living at every intermediate step, is known to be viable by the fact that it continues to exist.

Some scrappy things are alive; they can grow under their own power into something massive.

Some scrappy things are dead ends, just little odd things that never turn into anything else.

It's hard to tell the two apart, even for gardeners, at the beginning.

If you make a scrappy alive thing, at the beginning many people will say, "... that's it? That's the world changing thing?"

It's better to not have too much attention on it in that dinky phase, before it begins growing under its own power.

Death in a given context is a "game over" that we intuitively fear almost infinitely.

When you die, there are no more moves you can make, an infinite loss.

But we intuitively fear "death" (e.g. getting fired, having an idea shot down by a superior) in a given context as though it's a real death.

But of course that context is stacked inside of a nesting doll of other contexts, and death in one does not kill you in the others.

When you survive a thing that feels like death, you realize that it's not actually a death.

This is an empowering feeling, as a whole new universe of asymmetric moves open up to you.

If everyone else in that context is acting like failure is death, but you know it isn't, you can swing for the fences every time.

You'll look reckless or misguided to those others, but you know the secret and you're actually being more rational.

An example of this phenomena: for people who have only ever been employed in their adult life, the period where they're unemployed (e.g. before getting a new startup founded) feels like death.

The bank account keeps on ticking down, not up!

But once you realize that you're still alive and doing fine, you realize you were overly fearful of it for too long.

I kind of love that.

We're adjacent to Silicon Valley, but outside it.

Whereas Silicon Valley emphasizes tech and hyper growth, Berkeley emphasizes holistic approaches, organic growth; understanding of the implications in the broader socio-techno system.

Berkeley has associations with Unix, counter-culture, hippies, Christopher Alexander, etc.

Although the hacker ethic and things like Whole Earth Catalog grew out of Silicon Valley originally, Silicon Valley has matured and become more traditional as the tech industry has matured, and Berkeley keeps alive more of that energy.

When you hire specialists to a specific solution, it's hard to move beyond that.

A product can stand (mostly) on its own.

A feature only makes sense as part of a larger whole.

A feature doesn't have its own P&L, a product does.

To get precision you'll ask smaller questions.

Which is what is happening with econometrics, and most of science and society.

A version of the streetlight fallacy.

That is, you'll focus on asking questions that are easy to answer precisely, not the important questions.

The important questions will have answers that are fundamentally imprecise.

The right approach is not to ignore the important questions.

It's to get comfortable with uncertainty and imprecision

It rewards adding yet another jenga block to the tower of peer-reviewed knowledge.

And as we get deeper into the details as society, the jenga blocks keep on getting smaller and smaller.

More intricate.

Larger amounts of net new surface area to ever smaller amounts of net new volume.

Synthesis takes a step back, considers all of the facts and knowledge spread across various domains, and asks, "what are we to make of this?"

It identifies the human-understandable throughlines across various domains.

Cross-domain synthesis is basically not done in academia.

It can sometimes be done in industry, but only if it's got some plausible application (or at a place that is so resource-rich that it can tolerate employees pulling on random threads).

Other than that, it falls largely to hobbyists or authors of mainstream science books.

Which is why I'm so grateful for people like Alicia Juarrero and her book Context Changes Everything: How Constraints Create Coherence.

I mean, right there in the title is a mindblowing distillation: "constraints create coherence".

What an intriguing inversion, a figure-ground shift!

A distillation from a riff on the book, Apocalypse as Design Constraint:

"We think of constraints as limitations. We focus on the possibilities they bias against because these can be known in advance. What constraints bias a system towards is not always easy to see—and everything hinges on this fact."

A few reflections.

I am constantly amazed at the chaotic, frenetic energy in systems as small as a cell. The rate of molecules bouncing off each other just blows my mind.

Nature doesn't know what a cathedral is.

That is, in the famous tradeoff between a cathedral and a bazaar, nature is 100% bazaar.

Only humans think to build cathedrals, to approach problems top-down.

There is something beautiful and inspiring about complex behaviors that emerge entirely bottom-up, that are alive.

These kinds of emergent systems are totally alien to our minds, and yet they're the most natural, common thing in the world.

Highly conscientious people are hyper-aware of not breaking commitments they made.

Which also can be parlayed into being highly disciplined.

Of course, tricks like having always rules, not sometimes rules, help too.

And that I then tell people what I think their superpower is, and take a step back, allowing them to grow into it, to take risks and push their limits.

When you're maneuvering in a powerful person's blindspot, watch out, they might kick you.

You avoid touching the third rail, but invite your reader to if they choose.

You individually are less likely to get electrocuted and knocked out of the game.

You retain plausible deniability.

You don't force them to connect the dot, you allow them to.

An induced pull, not a push.

When a reader connects the last dot and has the potentially provocative spark of insight, it's electric.

They connect across a chasm of possibility.

The spark pulls them in.

They might become an evangelist for the idea.

Connect 9 out of 10 dots, to not be directly in danger.

Science fiction allows that kind of parallel too.

9 out of 10 dots connected allows you to dance with it, to sense it, so later you can have just a single piece fall into place in the listener's mind and go "oh wait, maybe I've been looking at it upside down."

That is, a "fool," a court jester, to help push back on the kayfabe of the organization.

Hire a Radagast to complement the Saruman CEO energy.

Try to tickle the company into novel perspectives by using safe-subversive techniques.

I've played this kind of role (unofficially) in many organizations I've been in.

This can be a dangerous role!

Doesn't matter who it is, as long as they're curious.

The virus of efficiency, optimization, rationality creates measurable value by eating away at the immeasurable value, the soul.

So I get into my flow state to self-soothe.

Which also happens to create value, directly and indirectly.

It can't escape that creator/participant.

Healthy communities have multiple voices active.

Blandness is like entropy.

The arrow of the universe points one way: from distinctive to bland.

The permission for people to do big, interesting, visionary, distinctive things has to come from the top in an organization.

If the top doesn't want it or allow it, then even if employees are capable of doing it, they won't try.

Only people can think.

"Because the data told me to" does not absolve you of responsibility for the outcome.

As organizations get larger, the machine gets bigger and more domineering.

The machine will put out numbers that compel you to a possibly dumb action.

"The machine told me to jump off a cliff"

"The UXR study told me to do it!"

Data is input, a source of potentially disconfirming evidence.

What ultimately matters is the informed judgment of calibrated people.

It doesn't stay open to surprising disconfirming evidence.

Resonating with other people, with the situation, with an ecosystem.

Indicators of quality of the feedback: how many steps are you thinking ahead, and does it loop back?

An OODA loop without the loop is bad.

Closing a feedback loop is about the ability to absorb disconfirming evidence from that loop.

A short feedback loop is better than a long feedback not-loop.

You have to have a series of nested loops, all of which close.

If any given loop doesn't close, you're in danger.

Long feedback loops with unclosed internal loops: you'll die before the long one loops back.

Short feedback loops with no surrounding long loops: you'll iterate into a corner or off a cliff.

N-ply thinking is multiple nested feedback loops at different scales.

Start with the smallest loop and keep on nesting longer ones on top.

Paint yourself into a corner, and then blame the corner.

"The environment is failing to conform to my prediction! What a jerk!"

It absorbs the cultural context it's in and becomes a particularly concentrated version of it.

As the tech industry has gotten more mature and more boring (more about banal first-order optimization), the PM role has also gotten more boring too.

It used to be more artisanal, about having a vision and creativity.

Now it's more about aggressive program management of optics.

That is, to spread resources evenly across projects, leading to at-best mediocre results across a large number of them.

However, spreading a diversity of acorns is good.

What's the difference?

Peanut buttering is about investing a moderate amount of resources in moderately successful projects.

Planting acorns is about very cheap bets that have asymmetric possibilities.

Later, you invest additional resources in seedlings in proportion to how well they are growing.

I said that across my time in the role I'd never not said what I thought, and had written down most of the things I believed.

I said that I had nothing to say that I hadn't already said.

Someone said about me: "Alex is the Kardashians of business strategy, everything's on camera"

The Toyota Management System saw that the factory line workers were not cogs, they were humans.

They were sources of situated insight.

Treating them as humans unlocked a feeling of empowerment and co-creation that led to not just a more empowering work experience but also more business value.

Luckily small things naturally have a more direct feeling of co-ownership, so people are willing to be scrappy for them.

It gets harder and harder to maintain that at larger scales.

If not, they will disengage.

If it feels like you all succeed together, a co-ownership, then people treat it as a more infinite kind of thing, non-transactional.

"Simply be more engaged" is a poor response to someone who is acting disengaged.

That's not how that works!

You have to develop a sense of shared ownership

But it happens very often in organizations.

For example, imagine seeing behavior that implies a lack of engagement from employees.

For example, employees canceling on interviews 12 hours before.

The natural reaction is to clamp down on the behavior and punish it more.

For example, "anyone who cancels an interview within 12 hours will have their manager and manager's manager cc'd".

The need for control ends up choking the thing harder.

Instead, ask what is the cause of this lack of engagement?

How can I create more space, more feeling of co-ownership and creativity?

How you react to disengagement will come down to where on the Theory X / Theory Y of management you fall.

(One of the worst-named frameworks, in my opinion).

Theory X: assume workers are fundamentally lazy.

Imagine someone crossing their arms in an X.

Theory Y: assume workers are fundamentally creative.

Imagine someone with their arms outstretched in the shape of a Y, cheering someone on.

Gardeners think in terms of systems.

Things are static.

Systems are always in motion.

Despite no coordination, the result is still coherent.

I used it as a metaphor for successive rounds of LLM token generation.

I learned last week that this game is a surrealist one called Exquisite Corpse.

If everyone is also participating in experimenting, then it's easy and fun.

Everyone is taking a risk, sticking their neck out, together.

If you do something embarrassing in front of the group, it's ok, the other people did, too!

This can build trust and non-transactional relationships

If you're doing it in front of people, it's terrifying!

It only takes one participant not being into it to ruin the whole vibe of an improv session.

That collapses it instantly from a something with to being in front of.

Midjourney succeeded in getting people to experiment and play in front of others.

No one judges you for how well you play.

Play is a fully infinite game.

Sometimes people judge you for the output of creativity.

"That's shit art".

Creativity is play with some kind of output.

Platforms like Instagram start off as a place to be creative, but become increasingly performative, for the likes.

Instagram is shallow play: play, but mostly for the likes.

For example, "Felt cute, might delete it later" -- Instagrammers keeping the optionality of deleting a post that turns out to be unpopular to maintain an illusion of perfectly popular posts.

Applying judgment to reduce a firehose down to an opinionated trickle of high-signal content.

Anyone who ever has chained together command line applications into a workflow will know this special kind of joy.

There are a lot of patterns from 15+ years ago that I wish people would rediscover the joy of:

Yahoo Pipes

Folksonomies

RSS

Composable, single-purpose command line applications

Anything Clay Shirky ever thought was cool :-D

The old thing doesn't go away, it fades in importance and relative strength, but some aspect of it will still be there even centuries later.

The withering effect of a powerful person looking at a seedling and saying "Why does this exist?"

20% time allows you to protect your own seedlings from bright sunlight until after they've grown.

Judo moves are like bending luck.

Look like a Saruman outside the team. Be a Radagast inside the team.

Inside the team be very authentic and candid.

Outside the team look very traditional and heroic.

The internals of a team is about ground truth, disconfirming evidence. The external of a team is about posturing, kayfabe, perception.

Who can you let inside? How can you make that as large as possible?

When a lead describes the hesitations of their actual decision to the team, it invites the team to see it as a fluid thing that can change, not a final, "this has been decided and now it's over".

It allows more disconfirming evidence to be shared.

It also invites co-creation of the decision, co-ownership, creating engagement.

They are often correlated, but not always. Similar to the oxygen/carbon-dioxide thing.

This is one of the reasons the overton window moves; people get habituated to it.