Scarcity and Abundance in 2025

I mean, what stage of the S curve is this?

It’s been a tough week for the AI haters; not only because of Ghibli Day (if you aren’t in the loop, congratulations for touching grass, but also open any Twitter feed and start scrolling) but also since Dwarkesh’s new book The Scaling Era is now out on Stripe Press, chronicling the oral history of AI so far. And the overall vibe this week got me thinking about an old series of essays I wrote at Social Capital - the Scarcity and Abundance series.

Those old Scarcity and Abundance posts were actually two different series I wrote at Social Capital, one in 2016 called “Emergent Layers” that iterated on The Innovator’s Dilemma, and a second in 2017 about positive feedback loops and Red Queen’s Races in environments of abundance. I think they hold up, 8 years later. They describe 2025 pretty darn well.

I’m gonna revisit them today. Now that we’re solidly in Agent S-curve somewhere - and you should read Dwarkesh and others for how that happened, not me - how do we start sorting out what resulting behaviour is actually new, what will continue to feel forced, and what’s next up that’s non-obvious? Three themes stood out to me, on reread, as plausible narratives for the next few years:

1. Affirming the idea of “light versus heavy businesses”, and how we think about software businesses and their profit potential

2. The eclipse of “Code as Capital” by a new metaphor that’s closer to “Code as Labour”

3. Agents that execute work outside of firms are more interesting to me than those that execute work within firms.

Let’s go!

Understanding Abundance

It’s been 8 years, so let’s do a refresher on the two series I’m talking about:

2017: Emergent Layers

2018: Understanding Abundance

The two series each tell a part of the story of how you get virtuous cycles that drive the cost of some critical input lower by orders of magnitude - Moore’s Law, Network Connectivity, and now the cost of reasoning and inference. In each case, consumers violently adopt a new S-curve when they are both overserved by what exists already (and like the new thing because it’s cheaper), and also underserved by the existing solution (and like the new thing because it’s better).

• Real S curves are pulled forward by market demand that violently appears, almost as if out of thin air. But at the beginning of the cycle, customers are jaded and “overserved” by existing products and incumbent businesses. Their value propositions are complex, and the customer has to consider many different factors in selecting a product. These are not the kind of starting points from which you get explosive new growth; something new has to happen.

• The spark that ignites an adoption curve is when a technological unlock (“newly abundant resource”) suddenly offers end users a new kind of product, sometimes a single killer feature, which is “disruptive” in the sense that it focuses on a single, newly practical value proposition compared to incumbents and ignores the “legacy” checkboxes. The decision to use the product is a quick “Yes or No” decision, not a check box ordeal.

  

• A virtuous cycle kicks off where increasingly diverse, highly specified solutions both a) pull infrastructure forward, and b) make consumer choice even easier for product adoption (you can just try things, and it’s a yes/no decision as to whether you want to or not.)

• A new “n of 1” company emerges for each of these cycles (approximately every 6 years?), that captures the scarce resource. It is not a mystery who the n of 1 company is. Everybody buys their stock, and they make a gazillion dollars.

  

• The interesting action lies in what people do with the newly abundant resource. It’s where the new behaviour is that really defines the era. For example, in a previous era of the web where Google was the n of 1 company, all of the interesting action was in Adwords and online advertising, where a new kind of “underserved customer” emerged with basically infinite demand to put the product do useful work, in markets like travel and ecommerce.

• Legacy companies don’t suddenly wither; they remain profitable for a long time, and are often put forth as “beneficiaries” of the new trend. But their product offerings can feel forced; the new tech compels them to overserve their customers even more, rather than deliver something truly new. (i.e. “now your email is sorted by AI relevance, and not by date!”) While it’s tempting to judge the new technology by these annoying intrusions, they are not the main story.

With that recap in place, we can pencil out the current state of software and AI-driven product roadmap development onto four quadrants: 1) the “Legacy” action that feels forced and tired, 2) the newly abundant resource that’s suddenly too cheap to meter, 3) the customers who can’t get enough of it, and 4) Nvidia.

Profit potential and the Red Queen’s Race

The first point to make here is to just ask the question: in these two quadrants where all of the action is happening, who is going to actually make money over the long run? There’s no doubt that real value is getting created for customers, but it’s less obvious which kinds of businesses are poised to capture defensible margin. Switching costs are just really, really low - which is, again, one of the bona fide hallmarks of the abundance cycle that I wrote about years ago. There are some whispers this week (that I won’t link to, because they’re just rumours) of various prominent AI startups being not-so-truthful about their revenue growth, which I think says something about the general mood: not that people are cheating, but that people are skeptical about the durability of any of these businesses.

AI today is a Red Queen’s Race all right: where you have to keep running faster and faster just in order to keep in place. The result is a harder challenge for both the pointy businesses (to stay ahead of consumer expectations) and the utility businesses (to keep deploying capital at immense scale, into things like data centres) to actually capture much of the value they’re creating, yet a hard cycle for either of them to escape.

(From 2017)

Amjad Masad of Replit, one of the shining stars of the Newly Abundant quadrant, shared on TBPN why even quick revenue is no guarantee of upward success anymore: the switching cost is so low between these different (and all excellent) products that practically all of the immense value being created is ending up as consumer surplus. Which is a great thing! Good luck picking winners on a valuation-adjusted basis, though.

(While we’re here, can we give a shout out to the Technology Brothers for creating such a great franchise, in such a short period of time? I’ve never seen anything like it. Hats off.)

We’re seeing the best example yet of “pointy businesses and utility businesses” being driven by a relentless cycle of specialization and differentiation, underlying infrastructure pull-forward, and consumer choice getting easier and easier. Amjad, understandably, talks about having to deliver “the whole burger” for customers as a way to remain sticky and differentiated. I expect the accelerating treadmill to keep accelerating for some time - I’ve never seen a Red Queen’s Race like this.

Peak “Code as Capital”?

Meanwhile, if we refocus our attention to the companies that already were big and advantaged, we see another classic “Scarcity and Abundance” theme play out predictably: the idea of the overserved customer getting even more overserved as a compulsive response from the incumbents.

From CNET

Very early on in technological S-curves, a narrative is often put forth that incumbents stand to benefit the most from this change - “this isn’t disruptive, it’s sustaining!” This was notably true for AI. There was a pretty conventional take, not long ago, that went: “AI is all about data, which the big companies have. And it’s going to be all about making existing products more contextually useful, which the big companies are best positioned to exploit.”

That’s not really what happened though, is it. The most beloved new products - ChatGPT, Claude, Cursor, Replit - are all startups. To the degree that tech giants’ versions do get adopted, as far as I can tell this is only from distribution advantages and not from any product advantage. Furthermore, (with some exceptions, like NotebookLM), the big tech giants with all our user data, all of the “contextual advantage” that was supposedly so important, compulsively ship a lot of annoying, forced features that no one really wants. (To their credit, the big tech giants have other pots on the stove that are way more beloved and successful, like Llama.)

Knowing all of our user context has not yet been the advantage people thought it was. Users are perfectly happy, it turns out, to bring the context they care about to the product, and then use the product as a focused tool to accomplish a specific task. For all the ink spilled about AI as an “orchestration tool” that knows my context and saves me time, more of the actual AI uptake has been focused on doing the end tasks. I expect that to continue.

If you try to put your finger on what exactly distinguishes the products that feel magical from those that feel like slop, you’ll notice something interesting. The pointless incumbent products - Genmoji or Text Summaries from Apple, “contextual” anything from Google and Microsoft, that kind of stuff - express an ethos of maximizing the value of a software asset of some kind. There’s a whiff of attitude where “the codebase is the capital”, and the point of all these AI tools is to keep drilling for undiscovered value in the asset.

Contrast this to everywhere I see people ravenously using new tools, like Cursor for coding, OpenAI Operator, or even fairly “basic” uses like lawyers using NotebookLM to summarize case documents in a way they can listen to in the car. There is no concept of an “asset” here; the value of the product to the user does not really depend on rich context, network effects, or some other obvious software incumbency. The software is just doing work, and the work is tangibly value-additive, even if it requires some human supervision.

It might be a bit too flippant to say, “We’re evolving from a mindset where the codebase is capital (the past few decades of software) and into a mindset where code is labor.” But this is a blog post, so it suits the medium. And it suits today’s energy: new projects and startups are writing a lot more code on the basis of “does this make me money now” (what Simon Wardley would’ve once called “worth-driven development”): the codebase is more like a workforce to be trained than like a factory line to be architected. You still want to put thought in it, but it’s a different kind of thoughtfulness. And you expect profit generation out of it a lot more aggressively than with patient capital deployment.

If you’ve gotten this far and wondered why the logo I put in the “underserved customers” box is Y Combinator, this is why - because new startups are arguably the most overserved by the existing paradigm of “your codebase is your capital.” As a new company, you don’t need, or want, “capital” as such! You want signal and growth, and if that can be accomplished by “labor” rather than by capital, all the better, if the numbers work. It’s no great mystery why the current batch of YC companies is earning more revenue than any other cohort, by far - it’s almost like a manufacturer suddenly discovering an ultra-low-wage and reasonably highly skilled workforce available for hire. You’re not going to overcapitalize this opportunity; you’re just going to put your foot on the gas and go.

This why I think it’s cute and possibly helpful to think of this period right now as Peak ‘Code as Capital.’ The true capital asset is becoming the big models and the chips; things you can finance, things you can rent. Going forward, we likely won’t romanticize individual companies or projects’ codebases as The Asset to the same degree, and I think is a double-edged but interesting development.

The Ronald Coase Theory of Agents

The last idea I have here is likely going to be the controversial one. I’m actually above-average skeptical about the ability of AI agents to radically increase “useful work done” inside of big companies - but in contrast, I think there’ll be wild and unpredictable new value created in the interstitial space between companies. Let me explain:

We talked about Coase’s Theory of the Firm the other day, and I keep thinking back to it as an old idea with fresh relevance here. If you don’t recall Coase’s work, “The Nature of the Firm”, it goes like this:

• Some kinds of economic activity gets naturally organized inside firms, whereas other kinds get naturally organized between firms.

• Markets generally work well, and strict classical economics would suggest that most productive work, over time, ought to be conducted by free economic agents and settled by market mechanisms.

• But this doesn’t always happen. There are some kinds of costs - the costs of searching for information, the costs of protecting trade secrets, the costs of coordinating complex sequences of events - that are more easily born within hierarchical companies than by free markets. Hence, “the boundary of a firm” spontaneously forms at the interface between work that economically makes sense to in-source versus outsource.

There’s one additional element here: the “O-Ring theory of productivity”. The O-Ring theory (named after the Challenger space shuttle disaster, where the failure of a single part ultimately doomed the flight) is about national economic development, but applies well to firms too: it argues that productive output is ultimately limited by the weakest link in a chain of economic activity, which in complex systems can be hard to identify. This idea fits really well with Coase. It says, look, firms naturally tend to in-source work that is highly susceptible to the O-Ring problem - the work that is complex and really value-additive.

Ultimately, the valuable work done inside companies is the work where human accountability is the primary bulwark against the O-ring problem. It acknowledges that complex decisions and accepted constraints need to have human owners, and that relationships between humans (notably, but obviously not exclusively, in reporting relationships) are ultimately the valuable stuff that companies are made of.

Okay, so, what does this have to do with AI agents? Well, my suspicion is as follows. One, combinations of AI and Human work could be notably vulnerable to the O-Ring problem. Their utility in executing a chain of tasks is only as valuable as the weakest link, very obviously. (If there’s one hallucinatory or badly-interfaced step anywhere that isn’t caught, you have to throw out the whole thing.) Firms have self-selected for solving O-Ring problems, over decades. That’s just what firms are.

So I think you could construct a bear thesis that the closer a workflow is to core to of an average business, the less suitable agents are to help, for reasons that are readily explained by Coase’s theory to the firm. Maybe not for the very best companies, who truly master their internal agents, but for the average company to whom AI Agents doing real work is essentially a kind of outsourcing. I’m not quite as pessimistic as that, but I agree with Tyler Cowen’s post that any hope for 10% GDP growth or whatever to emerge out of existing firms (which, even in a very disruptive environment, make up most of GDP) is too hopeful.

This sounds pessimistic about the value that can be added by agents, but I’m actually arguing the opposite is true - the most interesting new kinds of work are at the “periphery”. Work that takes place out in the market, where there’s plenty of skin in the game, and more importantly, that the economy has already selected, via Coase’s theory, as an environment where the lower-trust, higher-agency, direct-consequences kind of transactions were naturally sorted. The world of commerce and trading, payments, insurance, capital markets, what have you - these are all areas that have already evolved heavy lines of defence in dealing with unexpected surprise (and the antifragility to exploit it). AI interoperability is moving ahead by leaps and bounds with Model Context Protocol, which is an important step for an emergent layer of work from independent agent composability. If you want to talk “Emergent Layers” and look at what’s next, this is where I’d spend my time.

And, of course, (I think you may see where this is going) - we have an excellent playground all set up where you can speed run all of those experiments to find out which ones work:

We’re seeing real traction on the idea that “stablecoins are the natural way that AI agents are going to make payments to each other”, but there’s another way to think about the opportunity here that I wrote about in the Abundance series in 2017, which is that cheap blockspace is the natural place where independent agents are going to write state. This is the radically interesting thing: the fact that we now have cheap public state and execution environments (like Solana and Base) allows AI agents (call them “onchain agents”, if you like) to just be independent, stateless, functional actors that “can just do things”. Some of the most important APIs for agents are going to be ones like Thirdweb, where agents can just use a traditional Web 2.0 api to read and write to shared, neutral state (blockspace) without needing their own RPC or anything. The result is an additional layer of composability that augments MCP, since there’s a valid, permissionless, shared state to work with.

From Thirdweb MCP Server 0.1 Beta

The end result looks like a virtuous abundance cycle where on one side (the right side, below) you have onchain agents that are able to just do things and be stateless actors (because actual state, like account balances and what have you, is just stored onchain in a neutral third place outside the boundaries of any company.) And on the left hand side, you have consumers of their work, which could be people or frankly could be other agents, but in either case, the decision to consume an agent’s work is a simple yes/no answer: “Is this exactly what I want? Yes/no!”

Here’s how I illustrated that idea in 2017:

Back when I wrote this, quite a few people told me it was word salad hallucination, but I distinctly remember one person - who is now a Pretty Important Person in AI - taking the time to email me that the thesis was really insightful. We’re not there yet, but I think we can see it. Can you see it?

I’ll leave it there. Thanks for reading if you got this far, and I wish you a great time playing, building, and joyfully enjoying this strange new abundance in the software world.

---

If you’ve made it this far: thanks for reading! I’m back writing here for a limited time, while on parental leave from Shopify. For email updates you can subscribe here on Substack, or find an archived copy on alexdanco.com.

If you're first out the door, it's not called panicking

Four reads on the Trump economic narrative

So, I guess there are four possible scenarios here?

Scenario One: the Trump economic gambit is a good strategy, but they’re executing badly

Scenario Two: it’s a bad strategy, and they’re executing badly

Scenario Three: It’s a bad strategy, but they’re executing well

Scenario Four: it’s a good strategy, and they’re executing well

These are entirely fake axioms, but over the past month I think there are four distinct narratives out in the world that each support one of these quadrants. In the interest of learning in public (and just sharing what we think is going on live in real time - I wish I’d done this at the start of Covid), I’m going to try and write down my understanding of “what are the storylines going on here.”

These four scenarios are not mutually exclusive; the truth of the matter certainly contains elements of all four. If you’re following the news, then most of this won’t be brand new information or novel synthesis. It’s my attempt to organize my thoughts, with the full knowledge that a) I’m obviously out of my depth there, but hey, who isn’t; and b) new information is coming to light all the time, but we’re two months in so we do have something to work with.

It should go without saying - but I’ll say it anyway - that this isn’t investing advice and if you trade based on anything I write in this blog, you’re a bozo.

The starting assumption: we’re all working off the Miran paper

I’m making one big assumption with all of these, which is that the Stephen Miran paper A user’s guide to restructuring the global trading system, the more-or-less consensus inspiration for a possible “Mar-a-Lago accord”, is what most people are looking at as their crib sheet for reading signals. (Miran now chairs Trump’s council of economic advisers, so this seems like a reasonable assumption around the table.)

Miran is not a solo voice here; he’s a part of a recent chorus (including the thought-provoking book Trade Wars are Class Wars, which everyone should read, even if you don’t agree with it) that describes a bigger problem than American manufacturing decline. The problem is the US Dollar’s reserve currency status:

“The root of the economic imbalances lies in persistent dollar overvaluation that prevents the balancing of international trade, and this overvaluation is driven by inelastic demand for reserve assets. As global GDP grows, it becomes increasingly burdensome for the United States to finance the provision of reserve assets and the defence umbrella, as the manufacturing and tradeable sectors bear the brunt of the costs.”

Miran is describing a reflexive feedback loop:

1. For the past decade, global savings has disproportionately poured into US dollars, bonds, and equities - because of rate disparities (e.g. between Yellen’s Fed versus negative interest rates abroad), perceived safety & stability of US assets, and plain old chasing scarce growth.

2. This pushed up prices of US assets, which tangibly gave American companies a cost of capital advantage. Arguably more importantly, soaring stock prices gave company execs confidence to grow business (hiring; Capex) and soaring stock prices also made US consumers giddily consume a lot more stuff - much of which came from abroad.

3. This perceived outperformance of American investment opportunity reinforces global savers’ propensity to park their money in the S&P, rather than anywhere else. Rinse, repeat.

4. The problem is, this positive cycle accelerates our collective deficit spending.

5. At some point, this bubble bursts, badly. Everyone loses confidence in the US dollar, asset prices collapse, we can’t fund our military anymore, China steps in, it’s a disaster.

The Trump team believes that radical action needs to be taken to get us out of this cycle. Trump clearly authentically believes this; he’s been saying it since the 80s. That definitely puts it into the category of “too early is the same as wrong”, but maybe it’s finally right. This is also a team that really believes in the power of stories. They know how important it is to name things, label characters and plot devices, so that the voting public can track a narrative of their choice. What are the various storylines we’ve seen so far, and how are they going?

“3/3/3” (Good Strategy, Bad Execution)

The first story goes, okay, the first step to fixing this has to be fixing the deficit. Trump and the republicans have a mandate to make big moves here. We have a unique opportunity to freeze and reverse US Federal spending and bloat, Elon refactors the US civil service for a new generation, and the economy accelerates. If we can do those things, we can grow our way out of the debt problem and get the US back on better fiscal footing. This plan was memorably described in Bessent’s “3/3/3” plan: 3% real growth, 3% deficit, 3 million additional domestic barrels of oil.

(I’ll brush by the fact that one of these can’t work if the Permian won’t cooperate. But the goal of “we want cheaper energy and cheaper input costs” is valid and possible, maybe they’ll get barrel-equivalents from natural gas liquids, they could get it pseudo-domestically from Canada, etc, so let’s forgive it for now and move on, since it’s not the point of this post.)

The general idea of “grow revenues and cut costs” does actually work across a variety of situations! But the thing is, you have to do both of those things, fast. And so far there’s no sign of either.

From Brendan Duke on Twitter: so far, DOGE’s impact on outlays has been imperceptible.

During the earliest days of DOGE there was a tremendous “never bet against Elon, he is uniquely capable of getting this done” vibe out there that has, if not disappeared, at least attenuated. This blog is not about DOGE and I’m offering no takes on it here, other than stating what I think is obvious: other cabinet leaders (particularly Rubio) sound like they have had it with Elon, and Trump has started publicly walking back his authority and promises in recent days. On a pure dollars and cents basis it could still contribute real savings, and find a more moderate path to Jean Chretien style cuts. But betting markets are now forecasting that the overall impact will be modest.

Meanwhile in congress, things aren’t exactly going to plan on the “3% deficit” front either. On the “what promises are we willing to break” front, it doesn’t look like congress collectively feels the need or the legitimacy to break any. The wild card could still be military cuts - in keeping with the theme of “Europe has to hire its own security now”, plus in a technology transition, budget cuts actually help you get off the old tired defence primes and onto Anduril, so this might actually go okay. Let’s assume for now they do get some military cuts. Still, they are a long, long way off from anything resembling fiscal balance, and if they can’t fix it now, they’re not going to find new political capital somewhere to fix it later. It’s now or never; so I guess never. You’re left with “Monetize the US balance sheet” style gimmicks, which (after a one-time gold mark-to-market, neat, great) basically means selling off US federal land at 10 cents on the dollar, which would be a sad thing indeed.

Overall, this is the least intriguing of the four stories. New administration, new congress, new mandate to actually make progress on the deficit this time, and yet, it’s the same old story. If this were the only story, it’d be a letdown start to the Trump 2.0 economy. Don’t worry, though, we have three more to go.

“Tariffs made America rich” (Bad strategy, bad execution)

So far this has been the objectively funniest storyline of the four, although “funniest” has to be asterisked with “reputationally and progressively catastrophic.” It’s also probably the most authentic to Trump’s core character. He’s loved tariffs his entire life; he always thinks America is getting ripped off all the time, particularly by those damn Canadians. So this time around, a major pillar of the Trump platform is using protectionist tariffs to restore manufacturing back to America, in a way that simultaneously creates a lot of jobs, but also uses robots not people, since they were low paying jobs, which is why they went abroad in the first place. (???) Or something like that.

The purpose of these tariffs, as outlined by Miran, is dual-use: as a durable trade balancing mechanism that can be put in place long-term, but also as a short-term whack on the head, from which you can prompt countries one at a time to take steps that America wants, like currency revaluation. As the world’s largest consumer market by far, America ought to be able to divide and conquer its trading partners, reshore a bunch of domestic manufacturing, and win so much we get sick of winning, etc.

Not everyone is buying it:

Yes, if you stick with this policy for a long time, you might live to see the benefits - I am sympathetic to the idea of “if you can’t make steel, you aren’t a country” in some form. But the costs of this knee-jerk protectionism are immediate, and they sting, so you’d better be prepared to stay the course and stick it out.

If fully implemented, you’re going to raise something like 100 Billion dollars a year. Which is real money, but nowhere close to what you’d need for any kind of sustained major tax cut, let alone replacing the income tax altogether, as Trump initially spitballed as a way to drum up enthusiasm. And in exchange, you’re going to drag GDP growth by a half a percent or so, and in a particularly regressive way. (50 basis points of GDP, compounded annually, is an enormous cost.) That’s not a great trade. Moreover, you make the west as a whole less competitive, which should be the main point to all of this.

Living in Canada, I’m getting a fairly Canadian-skewed lens on what the tariff rollout has looked like so far. And from my vantage point, it looks like a comic book disaster. They’re putting on tariffs, but now they’re delayed; now they’re back, now they’re all mostly exempt, now the ones that hurt the US the most are on double. (Seriously - tariffs on aluminum, they day they’re declaring a national electricity grid emergency? Do any of them know how you make aluminum?)

The other problem with this strategy is Howard Lutnick. My working model of Lutnick - “the final boss of sell-side guys” - is that he is a genie of some sort, summoned out of a lamp, and determined to grant Trump’s wish of psychologically overcoming the jealousy of seeing the Biden guys actually succeed at some amount of manufacturing restoring. He has a supernatural gift to make things feel personal, not business.

You can imagine an alternate scenario where Trump ran the same tariff plan through a different messenger, and it’d be doing a whole lot better than it is now. What should’ve happened was Trump himself being chief bad cop, and a charming lead negotiator. In a recent David Remnick interview with historian Stephen Kotzin, they reminded us of Trump’s “Good cop bad cop” dynamic that he repeatedly set up in his first term: with our enemies, Trump cozies up as good cop, and lets his staff make the threats. With allies, he’s the bully, while his staff got productive work done. He could’ve run that here, and had it work great! Instead, Lutnick has just gotten everyone to hate him, and they’ve burned through a lot of credibility.

The other thing here is, just a regular recession, combined with some reshoring incentives, might’ve actually done a lot of help on the trade balance front. But these tariffs, if carried through, make inflation worse, make cutting rates harder, hurt your manufacturers in the short term as they have to pay more for all of their imported components… yeah I’m out on this one. Tariff policy, as carried out so far, officially rings the bell for “bad idea, worse execution.”

So those were the two “bad execution” categories. We’re failing to cut spending or get in fiscal shape; we’re failing at the “tariffs made America rich” McKinley story. What’s more interesting - and anxiety-provoking - is the framing in which it’s being executed well.

“If you’re first out the door, it’s not called panicking.” (Bad strategy, good execution)

This is the interesting one.

So, let’s go back to the principal Miran assertion, “we have to take action right now to get the dollar lower.” The problem with the current setup is that the US dollar and US assets have gone up when the global economy is strong (because we run a big deficit, other countries accumulate a lot of dollars, and keep them in dollars because that’s where they can invest them safely and profitably), but the US dollar also goes up when the global economy is weak (it’s the safe asset), so there’s no mechanism to get the dollar down.

Unless - and here’s where the Joker plan comes into view - the purpose of the plan is to throw things into complete chaos. Not garden variety recession fears, where people flee into the dollar. But chaos around “We can’t trust the United States anymore.” Sam Hammond has a thread around possible sequencing here; alternately, Yanis Varoufakis has a simplified version. We’ve already seen some early steps:

1. Europe received J.D. Vance’s message loud and clear that they can’t count on the US anymore, and have to massively raise their defence spending. Canada received Lutnick’s message that they can’t do high-trust business with the US anymore, have to vote Liberal now, and likely will spend on massive infrastructure projects to bring its natural resources to new markets. And so on.

2. The US announces an offer they can’t refuse: “Hey, we’re asking all big creditors to move into ultra-long duration treasuries that don’t pay any coupon; oh and if you want any military protection from us, you won’t say no.”

3. Then, the thinking goes, you get “a controlled retreat from the global dollar savings system, on America’s terms”. Other countries can’t just leave (it takes a long time to build an army!), but they want to: they don’t trust America anymore, they have their own stuff to spend on. That equals leverage to redraw a Mar-a-Lago type deal.

I think the big problem with this approach is that it leaves US businesses pretty vulnerable to a shock that is both bigger and less controllable than anyone in the White House is bargaining for. The problem is that the current system has two reflexive feedback loops in place:

• First is the reflexive loop of money-get-sucked-in: savers all around the world perceive US assets as both safe and growing; so they park their money there, which makes number go up, which makes their perception come true. (While earnings have gone up over the past decade, S&P gains are more than half from multiple expansion.)

• Second is the reflexive loop of Stock-is-up,-let’s-spend-money: the basic idea here being, the CEOs and management teams who make capital allocation decisions are primarily looking at their stock price as a signal for, “Are things going well, and should we invest more money back into the business?”. The perception (the stock price) meaningfully drives the reality (hiring, Capex), in both directions.

So, here’s what this means. If this plan comes true and there’s a big rotation of foreign money out of US stocks and into international ones (into foreign spending on their own domestic priorities), the US economy is not going to move like a market that was at equilibrium with the rest of the world - it is going to snap like a whip. This is the problem with reflexivity: if the flow of money stops coming in, and stock prices go down, CEOs suddenly get pessimistic (about their stock and also about all of the tariff uncertainty, etc), you see hiring projections go down, Capex go down, revenue guidance go down. And that can kick off a vicious cycle in the other direction. Reflexivity rewarded us in one direction with “the S&P basically goes up double digits every single year, and sometimes goes up by like, 25%”, but it could punish us in the opposite direction with an equally serious magnitude of outflows.

And, of course, we have a president who famously does not look at the stock market as his report card. Nope, not at all.

The other problem with this approach is you have to stop spending so much money domestically in order for it to work (see # 1). But that certainly doesn’t seem to be happening at a rate that will move the needle. So if you have a tanking stock market that meaningfully translates into sandbagged US growth, and you aren’t actually making any progress on the deficit, then other countries have just as much leverage over you - your ability to coordinate “controlled slow retreat from dollar-as-savings, but still have dollar-as-unit-of-global-exchange” just gets too fraught.

For this strategy to work, you have to really buy into the idea of “if you’re first out the door, it’s not called panicking.” (I hope you all know where that line is from without having to click the link.) I obviously understand why the Trump team refuses to sleepwalk into a scenario where other countries are dictating the terms and the pace of de-dollarization. But this kind of high-stakes negotiation will only work if you’re coming at it from a position of strength on all cylinders. Given that #1 and #2 are going badly, then I’d categorize this thread of the story as, “successfully pursuing a bad idea” rather than “badly pursuing a successful idea”, although I’d forgive you if you mentally categorize this as #1 instead. (In fairness, they have to go after them all at the same time; you don’t get to sequence them. Still, though.)

Anyway, all of this seems bad, bad bad. Is there anything we can find in our fourth quadrant?

“No Crying in the Casino” / Good strategy, good execution

It is unbelievably tempting in times like these to start listening to the doomer podcasts, start regularly checking the gold price, and buy into a theory that “it’s all going to collapse”. It’s good to be open-minded, but there’s a reason why the doomer cranks are historically wrong: the American economy itself is still pretty strong, and not only is it strong, it’s pretty resilient compared to any other alternative. The stock market is not the economy.

Moreover, let’s go back to Trump and Bessent’s stated goals: for rates, the dollar, oil, and inflation to all be going down. Well?

(From a longer post from Gavin, a perennially thoughtful and reasonable source.)

Sorry if this sounds obtuse, but… what if they’re getting it? We got our stock market correction, particularly in tech, which is down anywhere from 15 to 35% from the election highs, but no one thinks that any of these companies are actually in any sort of danger. Market stress has nudged some people into bonds; their focus on the 10-year is going fine.

Meanwhile, Europe has answered the call, loud and clear, to not only fund their militaries but also to move past the “Debt Brake” era and grow into an alternate western consumer / builder superpower again. We’re getting some overseas investment into what they want to see (American manufacturing production) rather than what they don’t (the S&P).

None of those things, by any means, are sufficient to get America out of the trouble that it’s in. But we are still early into this term, and we have to remind ourselves that there’s a lot we’re not seeing as of yet. That could mean things are even worse than the headlines suggest, but there is no shortage of essays and takes to that effect. I’m not smart enough to say I am following it any better than the average person who follows smart people on Twitter. You’re allowed to say it’s hazy!

Okay so, what do we think?

If I were to weigh all of these together, I think the one aspect of this you can safely bank on is that we’re not seeing sub 3% deficits any time soon. Every other story, like tariffs or equity rotations or the alchemy of rates, has too many unknowns for me. But congress isn’t likely, or able, to pull off a Hail Mary or hidden ball trick the way the executive branch might. And, as much as you should never underestimate the guy, the range of plausible DOGE savings is considerably smaller than it was a month ago. Still could help, but it won’t carry most of the water.

Again, not investing advice. Stay safe out there, get lots of opinions from a wide range of sources, get primary source material when you can; and stay curious. Like the 70s, when we got to test every economic theory at once, we’re about to get some real data on how political and economic power actually work now, about which we could only grasp and speculate just recently. And maybe we’ll look back on this period of time and all of this will be a footnote to something even bigger.

---

If you’ve made it this far: thanks for reading! I’m back writing here for a limited time, while on parental leave from Shopify. For email updates you can subscribe here on Substack, or find an archived copy on alexdanco.com.

Wine tasting with Claude Shannon

Why we can’t resist a grain of truth within a crazy argument

Before we start: on April 16th, I’ll be hosting an Interintellect Salon on Thomas Mann’s great novel The Magic Mountain, which is briefly mentioned in this post. I’d love to see you there - you can register for the salon here.

---

There’s some counterintuitive but deep intuition that things can feel higher signal when they are 99% false but 1% true, versus it they are 99% true but 1% false. A grain of truth within a crazy statement can seize our attention more powerfully than a reasonable argument with blemishes here and there.

There are many variations of this: like how a tiny pinch of salt brings out more flavour in a dish than a larger dose. Or how the 1% of pro wrestling that’s off-script brings the whole show to life, whereas a football game that’s 1% fixed by refs (but still 99% genuine) would be immediately declared fraudulent.

Sometimes homeopathic doses do have stronger effects than conventional ones; at least in the world of information and meaning. I think you should be curious about why this is, if you want to be information-literate in today’s world. You want to be in tune with how powerful that 1% signal is, but also have the presence of mind to not fall for things you shouldn’t.

To get to the bottom of this, in this post we’ll spend some time on what exactly information is. We’ll learn it through the kids game of 20 Questions, and the grown up activity of blind tasting wine, to get some intuition around why those grains of truth disproportionately draw our attention.

Information is a verb, not a noun

A lot of people today don’t know who John Perry Barlow is, which is a cosmic oversight. His life is more interesting than I can summarize here, but as a young man he became the Grateful Dead’s lyricist, and headed up the contingent of Deadheads who helped shape the early internet. He cofounded the Electronic Frontier Foundation in 1990, and his writing is closely associated (and often confused with) Stewart Brand, and with famous lines like “Information wants to be free”. (With the eternal clarification, “free as in speech, not as in beer.”)

Buried inside one of Barlow’s essays on digital property rights, there’s a profound paragraph that I’ll copy here in full:

Information Is a Verb, Not a Noun. Freed of its containers, information is obviously not a thing. In fact, it is something which happens in the field of interaction between minds or objects or other pieces of information. Gregory Bateson, expanding on the information theory of Claude Shannon, said, "Information is a difference which makes a difference." Thus, information only really exists in the [delta]. The making of that difference is an activity within a relationship. Information is an action which occupies time rather than a state of being which occupies physical space, as is the case with hard goods. It is the pitch, not the baseball, the dance, not the dancer.

We’d joke today that “Barlow is a just Bateson wrapper, and Bateson is just a Shannon wrapper”: a lot of roads lead you back to Information Theory. The definition of Information as “a difference that makes a difference” took me years to really understand, even with alternate phrasings like, “Information isn’t what you say, it’s what they hear”. I finally wrapped my head around it after reading Simon DeDeo’s excellent paper Information Theory for Intelligent People. I absolutely love this paper. I find it so clear and intuitive, I find myself rereading it every six months minimum, and I’ll draw from it extensively in the following explanation.

“I’m thinking of a person”

My older kids have reached the age where we can play Twenty Questions at the dinner table. (We play it very simply: one player thinks of a person, and the others have to guess who from Yes or No questions.) After playing for a little while, you learn that the main way you can “get better” at the game is to know something about the other player and tailor your questions appropriately. You can talk about an optimal strategy for each kid: for our older one, her schoolteachers are a promising lead; for the middle child, try asking about her classmates first.

This is the moment where Information Theory enters the picture. I wrote a post a few years ago that walks you through the math (it’s actually pretty approachable) so I won’t repeat it here. All you’ll need is a bit of intuition around probability distributions, which 20 Questions provides for us handily.

Let’s do an ultra-simple example: p(Mom) = 0.25, p(Dad) = .25, p(Big sister) = .25, p(Little sister) = .25. You can quickly work out that this distribution, P(x), takes exactly two Yes/No questions to solve. Now let’s suppose Dad’s ineligible, and Mom’s twice as likely. So we have a new distribution Q(x): q(Mom) = .5, q(Big sister) = .25, q(Little sister) = .25. You should intuit that this should take less than two Yes/No questions on average, because if you guess “is it Mom” right off the bat, half the time you’d be right. Shannon’s equation lets us calculate, from known probabilities of Q(x), that our average question tree length will be 1.5.

Once that’s comfortable, let’s go ahead and play a real game. (We’ll reuse our P and Q variables.) For child one and two, we’ll define distributions P(x) and Q(x): the set of people that they routinely select for you to guess, and the relative likelihood of each. We can say a few things about these distributions:

• The older kid knows more people and is less predictable. We might expect the average question tree length to take, say, twelve questions for the older kid versus eight or nine for our middle child, on average. So we can say we expect P(x) take 3.5 more guesses than Q(x) on average. Each yes/no question resolves one bit of information - a unit of resolution of uncertainty, in base 2. (It’s base 2 because we’re asking Yes or No questions, and we assume that at some point we reach a satisfactory level of yes/no detail that no further questioning is required. This is called “Coarse graining”, and DeDeo’s paper has a great explanation of it.)

• The older kid will reliably think of a child ~50% of the time and an adult 50% of the time, so we can meaningfully talk about 1 bit of uncertainty to Child/Adult that we must suss out with our questioning. In contrast, our middle child usually thinks of a kid rather than a grownup, so there is less than one bit of uncertainty there. (Go back to our simple example earlier if it’s not obvious why.) If she thought of a kid every single time, which is close to what we empirically observe, then there’s actually zero bits of uncertainty in Child/Adult, so we could safely skip the question. We might skip straight to a finer-grained question: “Are they in preschool?”, a poor strategy for the older kid but optimal for the younger.
  

Why does this work? As the guesser, the fact that we know something about her map of the world - how big it is and how it’s coded - means our guessing process can take fewer questions on average. It has less uncertainty to resolve. That quantity of “uncertainty-to-resolve” is called H(X), or Entropy. It is not an absolute quantity, it is a Delta; and it’s measured in Bits. This is very much in the same world of Bayesian probabilities, and managing priories; but I find the Shannon lens gives an important emphasis on “Uncertainty-to-resolve” so we’ll keep going with it.

Surprise!

Now, let’s imagine we’re playing the game with a blindfold on, and you don’t know which kid you’re playing against. Pay close attention here because this is the critical step for understanding what Barlow and Bateson are getting at with “Information is a verb, not a noun.”

So in this blindfolded version, we don’t know if we’re interrogating P(x) or Q(x); but we must commit to a questioning strategy at the outset. So let’s say we choose a questioning strategy that’s optimized for the older kid’s probably distribution P(x), but then over the course of our questioning we discover that, actually, we’ve got the younger kid. Rats! We were inefficient with our question tree. We can talk about the additional work we must do of asking questions when we were expecting one distribution, P(x), but actually encounter another one, Q(x).

Note carefully that these additionally incurred questions are not because the distributions are contain different amounts of latent uncertainty. It’s because our questioning itself had been non-optimal, and now has to be recalibrated. That additional uncertainty has a formal name, “Kullbach-Leibler Divergence”. And it measures the extra uncertainty, in bits, that we incur in the discovery that we are dealing with a different distribution than we expected, and it’s a signal to focus our effort in a new place. The everyday name for this is “Surprise.”

These moments of surprise are when we actually learn things. It’s the moment when you realize that your map of the world - your internal model of the current situation - does not actually match the facts on the ground, and we need to switch up our map. Whatever answer surprised us with, “this is actually Kid 1, not Kid 2”, was a difference that made a difference. To spell it out: it was a difference in observed versus expected values, that makes a difference in our selection of coding tree, which can be measured in bits of newly appreciated possibility associated with the new uncertainty.

I’ve written about this before, in a post called The Audio Revolution. The relevant part was about feed-forward information processing, and how most of the time, the world we perceive is actually a pre-populated model of objects and spaces from our working memory, which we then “error-check” against a low-bandwidth sample of real sensory information. When you’re in a familiar environment, like your living room in your house, you draw on a richly defined, pre-populated environment of what everything in the picture is (call it P(x), and a well-rehearsed strategy for what “questions” you ask of the world. But if you walk into your house and something is amiss - say, the TV is missing - that is a difference that made a difference in what your model of what’s going on, and what questions you now ask of the world in order to figure out what’s going on.

This is why it’s actually incorrect to say “They said something surprising.” They just answered the question. What actually happened is you HEARD something surprising. That’s what information really is. Information is not the baseball (the kid’s answer), it’s the pitch (the impact of the answer on the model that was perceiving it.) It’s not the dancer, it’s the dance.

From Itti & Baldi, 2009: Vision Research

As it turns out, our brains are optimized to care about surprise, and we can measure this scientifically. You can show someone a movie and calculate, for each pixel on the screen, how much uncertainty there is in each pixel over a time series. And you might think “our eyes are probably drawn to the pixels where there is uncertainty”, but that’s not what happens. Maximal uncertainty is actually like the “snow” pattern on an old TV - it’s not visually interesting. What our eyes are actually drawn towards is, guess what, KL Divergence. Our eyes follow the pixels that are surprising. Because that’s where the information is!

In our 20 Questions game, these areas of “surprise” where our eyes are drawn are analogous to whatever bit of information makes us realize, “We have the younger kid, not the older kid.” That little bit of information is probably a small thing, but it’s a small thing that makes all the difference. And that’s where our attention is drawn.

Although Twenty Questions is a great starter example, it’s not very realistic. In real situations, we don’t always get to directly interrogate the Ding an Sich (Kant’s Thing In Itself); we’re dealing with interpretations, representations, unreliable signals, and noisy channels, from which we have to extract and recreate a signal. (The one way in which 20 Questions with kids is realistic is that they are not totally reliable narrators! Their models of the world sometimes differ from ours, and knowing which answers to trust is part of the meta-game.)

Freshly Cut Garden Hose

As it happens, several friends and I have recently gotten into a more grownup version of 20 Questions: blind tasting wine. It adds the dimension of noisy, unreliable signals, and even better illustrates KL Divergence and Surprise.

Blind tasting wine is a fun challenge, and can be absolutely humiliating. It’s a lot like investing - it’s easy to find yourself on a hot streak and think you’re smart, but very hard to actually be skilled at it. But you can work at it and get better. We talk about “training your palate”, and a lot of it is learning to recognize discrete smells and tastes, but it’s just as much a mental training exercise for the general skill of how to sift through and resolve uncertainty.

In moves like Somm, or other romantic representations of wine tasting, you get an impression that candidates are solving an elaborate 20-dimensional puzzle:

• What colour is the wine, and what does that tell you? Is it clear or opaque? A red wine you can look straight through invites several different lines of inquiry (Thin-skinned grape variety? Made it Italy?) and dissuades others (It’s probably not an Aussie shiraz, or an Argentinian Malbec.) A white wine that’s copper coloured might put you in pinot gris territory, but specifically from France, not Italy. But is that really copper, or is it golden?

• What does it smell like? Very perfume-y white wines (Viognier, Gewurtztraminer, Riesling) can be triaged from those that aren’t characteristically aromatic (Gruner, Chablis, White Hermitage) to get you in a narrower zone.

• How hot was it where this wine was grown? Was it grown in a cool place (it’s nervy, tart) or a warm place (jammy, rich, high alcohol)? What about a hot year in a cool place?

• What kind of texture does the wine feel like when you taste it? For red wines, is there tannin (the astringent texture of eating walnut skins), acid (makes your mouth water), tons of both (Italy?) or more one than the other (Pomerol? Beaujolais?) Same for whites - is it acidic, bitter, or both?

• Is there oak on the wine? Is it American oak, which tastes like coconut Lacroix and dill pickle, classically used in Spanish reds? French oak, with vanilla and spices that suggests the wine is expensive? Is the oak new (the taste is obvious) or old (barely any taste, just texture?)

• How old is the wine? Is the fruit dominating, or has it faded and evolved into other tastes like leather? Is there any orange or brown tint to the wine?

• Are there any “giveaways” that immediately hone you in on something? Like black pepper (think Syrah), celery (Gruner), green pepper (Bordeaux varietals), tomato leaves (Sangiovese), cat pee (New Zealand Sauvignon Blanc), gasoline (Riesling), tennis ball can (Australian Riesling), roofing tar (Nebbiolo), peach yogurt (Zin), Icy Hot pain relief cream (Australian anything)…

• And finally, I’m saving this one for last because it’s “the most information” but often the most confusing and least helpful - what does the fruit taste like? Red fruits, black fruits, tree fruits, stone fruits, the flesh of the fruit, the rind of the fruit, the pith? How ripe?

As fun as these “grid tasting” portrayals are, they make it look easier than it is. (It’s a bit like watching poker on TV, when you can see the cards.) In practice, you’re not dealing with clear signals: you’re dealing with broad correlations and faint hints. Any one data point (“raspberry fruit”) could point you in many different directions. Moreover, you aren’t always so sure you’re really tasting something, confusing it for something else, or just imagining it. Taste is notoriously adaptive to the power of suggestion. (Curiously, the one sense I find people chronically under-use is their eyesight - how the wine looks will not fool you, at least not nearly to the same degree as what you think it tastes like from minute to minute.)

Wine can be tasted “double blind”, where you know nothing about it at all, or “single blind”, where you know some information (e.g. what grape, or what region) and you have to figure out the rest of its characteristics. The standard things you’re trying to figure out are, what grape(s), what region, what year, and if you’re particularly ambitious, what producer. While Double Blind tasting looks more impressive, it’s generally thought that Single Blind is the more valuable exercise for learning how to taste - you aren’t floundering around in the dark; you can actually control for some variables while you interrogate others.

Jensen-Shannon Distance

So, when we think back to what we learned in 20 Questions, wine tasting is the same idea: particularly the concept of KL Divergence and “Surprise”. Wine tasting is all about pulling out maps of what you think the wine might be (P(x)), while remaining alert to signals that don’t fit (and suggest you’re looking at Q(x)).

The temptation, particularly for new tasters, is to approach the wines inductively. Let’s do an example! You try a red wine; it’s solidly built with grippy tannins, and old world oaky tobacco-leather taste. Then, like a laser beam, you pick up cassis: black currant fruit. Your mind immediately goes to Bordeaux. Yep, that’s delicious. You try to focus on where in Bordeaux this could be - left vs right bank. Black currant suggests cab; it really reminds you of a Margaux you had last month, so let’s guess that.

The wine is revealed, and it’s not Bordeaux at all. It’s from Rioja, in Spain. Where did your line of questioning go wrong? You can retrace your errors. You were initially on a good path with the oak, tobacco, leather, and black fruit. But that little signal of “cassis” drew away all of your attention, and made you pull out the Bordeaux map. And once you were reading from the Bordeaux map, you missed some pretty big clues.

First, the wine has a screaming taste of dill pickle, which should tell you American oak, and therefore you’re in Spain not France. Second, there’s no trace of green pepper or other pyrazine taste, which doesn’t strictly rule out Bordeaux, but certainly suggests you consider alternatives. And third, tasting it again, that isn’t really black currant, is it. It’s more like black plum. Close, but not the cabernet signal you’d initially thought it was.

(If you like these tasting note differential sheets, they’re from the excellent book Vignette, by Jane Lopes)

The moment you picked up that cassis signal, you were “Surprised” in a particular direction. You interpret it as “Get out the Bordeaux map, that’s the distribution we should be working off.” And then in asking questions about P(x), you got some fuzzy answers that suggested one place or other. But we were wrong about the wine, and we were using the wrong map. The KL Divergence of Tobacco overwhelmed our ability to look around the rest of the picture. We were surprised by one thing, and it blinded us from being surprised by other things.

The failure is that you immediately got into a headspace of “I recognize this.” Inductive reasoning is dangerous, because you’re telling yourself a story about what your senses are telling you, and then you spend all of your energy focusing on that story. So you may miss some big obvious hints.

What we are supposed to do when we blind taste is to resist this temptation to leap to a hypothesis. The right way is to approach it deductively. You look, smell, and taste the wine, and just try and describe it as non-judgementally as possible, gathering information. Then, based on all of your observations, you use logic and theory to rule things out: “What can this wine not be?” Is going to fool you much less of the time than “What do I think it is?”, because you’re not latching onto an inductive hunch. As you do that for a while, you build up for a theoretical case for what this wine is all about - the weather, the winemaking, possible grapes, possible terroirs. Only then are you allowed to float some hypotheses for what the wine is.

Similar to KL Divergence, another term you can take home from this post is “Jensen-Shannon Distance”, which describes how much information you get about the distributions (P(x) vs Q(x); or, “typical Bordeaux vs typical Rioja”) you get from each question you ask, like “Is there any pyrazine?”. The units of JSD, once again, are in bits - how much information you get about the final answer once you know the identity of the “player” (which could mean the grape, the region, or whatever taste profile is in your mind.) For more on JSD (including the important topic of “mutual information”, which I didn’t get to cover here, read the DeDeo paper.)

That’s not to say that a deductive approach leaves you completely invulnerable to false trails. But by analytically working through a grid of characteristics and working through what the wine isn’t, sooner or later you’d encounter signals like “No green pepper” and “Yes dill pickle”, both of which are signals that you are not dealing with distribution P(x), you might be dealing with a different Q(x). And the earlier you can catch these big signals, the more likely you are to see them at all. Because once you get blinders on and you’re going down the path of “Cassis!!”, all of the other signals are just going to get force-fit into your reading of P(x) and you’ll miss what’s actually going on.

It’s never Lupus

There’s a line in medicine, “If you hear hoofbeats, think horses, not zebras” that’s used to illustrate Baysean probabilities. Learning how to appropriately weigh prior probabilities is an important part of learning how to think well. (The classic Bayesian intro question is: “Alice is a quiet person who loves organizing and doesn’t like to be rushed. Do you think Alice is more likely to be a barista or a librarian?” And the right answer is barista, simply because there are many more baristas than librarians out there, and the special information you gained about Jane is small relative to the employment prior.)

What I hope you’ve gotten an appreciation for here it that tiny bits of information, assembled amidst otherwise broad and predictable distributions, have a powerful effect on where we focus our attention. A great example from TV is in House, when tiny details throw him down paths of being absolutely certain he finally has a case of Lupus, but it never is. Now you have a better sense of what’s going on: that one tiny detail that stood out to you seizes a disproportionate amount of your attention, because it tells you “New KL Divergence just dropped, you have work to do”. You then pull out a new map (a new question strategy, in 20 Questions; a new grape or region hypothesis in wine tasting) and from there on, the main thing that matters is interpretation of that new hypothesis. From the little black currant clue that sent us into Bordeaux territory, we then found all kinds of signals took us to Margaux as opposed to St Julien or whatever, even though the broader data set did not actually support Bordeaux all that much.

It reminds me a lot of the book The Magic Mountain by Thomas Mann, where the intellectual spine of the book is an ongoing dialectic between two Italians. One of them is a bourgeois named Settembrini who’s comical, but basically sensible in what he advocates. The other guy, a character named Naptha, is a Jesuit communist warmonger with crazy arguments (among other things, at one point he argues forcefully that heliocentricity is just a passing fad), but with undeniable little grains of truth sprinkled in that are unbelievably compelling, and kind of right. Reading the book, you get fixated on those grains of truth and a whole world of legitimacy gets constructed around them - even if most of his raw content is insane. You can’t help but be fascinated by him, and he somehow wins every argument. He reminds me of RFK a little bit.

It is hard for us to resist these grains of signal, because they are what signal IS. And it’s equally hard for us to step back and sort through what else we’re perceiving, because they’re not signal, unless we really train ourselves to set aside our inductive instinct. What you can do is remember Bateson’s definition of information: a difference that makes a difference. Information is a change in uncertainty-to-be-resolved, and it happens to the receiver.

---

If you’ve made it this far: thanks for reading! I’m back writing here for a limited time, while on parental leave from Shopify. For email updates you can subscribe here on Substack, or find an archived copy on alexdanco.com.

10 Predictions for the 2020s: Midterm report card

In December of 2019, which feels like quite a lifetime ago, I posted ten predictions about themes I thought would be important in the 2020s. In the immediate weeks after I wrote this post, it started to dawn on everyone - gradually and then suddenly - that the Novel Coronavirus found in Wuhan was not going to be a normal news cycle. And so, for a few years, any kind of forecast you could make about the 2020s had to take a back seat for current events.

But today, we’re at the halfway point through the 2020s, and not to downplay Covid or anything, but we’ve actually returned to something like our old story arc. So let’s check in and see how we’re doing.

For readability, I’m not going to quote the whole text of each prediction - but I encourage you to click through to the original piece, which I have left untouched, as I do for all of my posts.

#1: Enterprise software in the 2020s will replay Softbank’s Capital-as-a-Moat disaster of the late 2010s.
Grade: B-

“The issue is that AI is really expensive. It’s more compute intensive, more data-intensive, more resource-intensive in general. The rewards will be immense, but AI businesses will face heavy fixed and variable costs. So there’ll be a huge advantage to being the market leader; more so than in the previous generation of SaaS businesses.” … “The scale advantages in AI businesses may simply become a new frontier for capital-as-a-moat funding strategies, and there’s a real danger that enterprise software could turn into Softbank 2019 v2.0. That’s life if you’re going horizontal, but it’s dangerous if you’re going vertical.”

So, I made two parallel predictions in the original call - one of which approximately played out, but the other didn’t.

The part I got right was “AI is going to be really compute and capital intensive, and capital-as-a-moat will continue to be a thing.” We absolutely did get this, and we’re now a couple of years into a giant Capex cycle for that exact purpose.

I got two things wrong, though. First, I was imagining this in a context of application companies themselves having to raise and spend all of this money, which isn’t what happened - instead, it’s the frontier models and the cloud giants burning capital, while application companies like Cursor and Replit reap the benefits. Second, I was wrong about this being a vertical market story - what’s actually happened was AI workloads taking over the most horizontal slices possible, like search, writing code, customer support.

And then there’s one last bit that I’m going to cheat and give myself half credit, which was the call about lots of term structure being required to fund all of this. Well, I can’t say I got the call right exactly, but the invention of the GPU-backed loan (or whatever the structure variant) has got to count for something.

#2: There will be no major form-factor that supersedes the smartphone. The phone is it.
Grade: B

“If you went forward in time to 2029, you’d be surprised that the phones are more or less the same. The 2×5 inch glowing glass rectangle will remain more or less similar as our common interface with the internet and with the world. Nicer in some ways, and they’ll have some genuinely cool AR features, but other than that? We figured out the phone. It’s gonna stay put now.

What’s inside the phone will be radically different, for sure – the “re-nationalization” of technology into Western and Eastern tech stacks will place some really interesting pressures on the hardware under the hood. It’ll be especially interesting to see what happens to the Android ecosystem as these pressures intensify.”

Correct, although this was a pretty boring prediction. When I think back to the time of writing this, there was some real hype around AR/VR and new form factors - the Oculus Rift S shipped that year; the rumour mill about Apple shipping smart glasses had kicked into rhythm. And there was a narrative around AirPods being the true gateway into native AR that was a fun thesis. But all in all, the phone was clearly staying put as the device and it’s clearly stayed that way.

Rereading this, I’m a bit disappointed that my hope for cooler vision features out of the iPhone haven’t materialized. Maybe we’ll get them in the back half of the decade, as Apple flexes its on-device computing advantage for making AI useful in the real world. I hope so, that’d be neat.

#3: A tale of two cities: the looming feud between the SF and NYC tech scenes
Grade: F

“In the next decade, watch for this rivalry to take shape. … Each city’s tech scene will come to resent the other’s more and more, and allegiances will be hard to shake: NYC will resent SF for its endless reservoir of technological and tech-cultural capital; SF will resent NYC for being a place people actually want to live. But most importantly, both tech scenes will come to resent each other for being more alike than different with each passing year.”

Our first F. Clearly failed here; this didn’t happen, it’s not going to happen. I will note that there was that little storyline about “Miami vs SF” for vaguely similar reasons, but doesn’t count as redemption. I’ll leave it there; that prediction was a zero.
(If this is happening and I just don’t know about it, please let me know!)

#4: At some point in the 2020s, the following headline will get printed: “Crypto Finds its Killer App: Guns”
Grade: D / B

“Over the next ten years, after our current hangover gets truly washed out, a new one is going to emerge that’s a lot stronger, a lot darker, and a lot more true to the original vision than many people would like to admit: crypto as a part of “Dissident Tech”. … Expect the crypto community to retrench and double down on a core value proposition: crypto as a tool for political and societal dissenters. This means people who don’t really care what the price of Bitcoin is; they actually care about it being uncensorable. That means people who are doing illegal things, it means people on the wrong side of governments, or it could simply mean freedom-oriented people. But the narrative will turn really sharply once the core value proposition embraced genuinely becomes ‘be un-censorable’, as opposed to ‘get rich’.”

Ok so, there were three parts to this prediction and I get some pretty divergent grades when you break them down.

The first guess was, “Crypto winter for quite some time; it’ll come back as Dissident Technology before Number Go Up technology.” F! Or at least, like, D minus. That is not what happened.

The second guess was specifically around ghost guns and payments. There has been some story here; not the literal headline I’d imagined, but the Supreme Court hearing around regulating gun 3d printing under firearm rules did key into crypto payments as part of that grey market ecosystem.

The third part, though, I think I nailed, and that’s on the topic of privacy itself. There is zero doubt that privacy and free speech has come to the forefront as a major global issue for our time. And it’s pretty obvious to anyone paying attention that the origin of this energy came from in and around the crypto community, from the Free Ross movement to the un-cancellable movement. I don’t think it’s a coincidence at all: the social ethos of that community was a critical incubation ground for what’s now a full-blown reactionary movement, and I’m not sure it gets going in the same way had crypto not been around in 2021-2022.

So, I’m going to cheat and take a hybrid midterm grade of “D on the literal call, but B on the broader story.”

#5: Higher ed: undergrad will stay the same, but grad school will get blown up
Grade: C

“I think that the narrative that American higher education is going to collapse in a matter of years is just totally false. There are so many structural forces holding the system in place, and (for most middle or higher income families) so much professional and reputational risk associated with not going to college, that the undergraduate experience isn’t going anywhere.

Grad school, on the other hand, is a totally different story. I’m going to write a whole newsletter about this in the new year but for the time being, I think both “classes” of grad school – professional degrees and research degrees – are going to be disrupted pretty dramatically, and pretty soon.”

Mixed bag here. On the one hand, there was this trend five years ago around “ISAs fully replacing college for most people” that I thought was nonsense, because it completely misunderstands the societal role of college. Yes, it has quietly become more acceptable for specifically motivated and capable people to skip straight to building things, but for the most part, the undergrad education pillar is unchanged and it probably stays that way for a long time. I think that was a pretty easy call so it shouldn’t contribute much to the overall grade on this one.

Grad school, on the other hand, remains ripe for a wave of disruption that we haven’t seen play out yet. Two months following the original post, I wrote down my thoughts on the graduate student industrial complex in Can Twitter Save Science?, and I remain a big fan of that piece as a coherent expression. And the big Trump shakeup of the NIH budget for university overhead, if it goes through, will have a massive impact on how these places actually function.

And it does look like the shine is starting to really come off of the MBA-to-consulting pipeline, among other job pathways:

On the other hand, the guess I’d made around the big trend being “Companies going all-in on post-undergrad vocational training programs” doesn’t look like it has legs, at least not that I’ve noticed. Still, these are early changes and the system still has a lot of inertia that’s still in place. So overall I’m giving myself a C here: we haven’t seen the rate of change that would be necessary for this prediction to earnestly come through by the end of the 2020s. But some of the trends are right, so we’ll see in five years.

#6: There will be a major speculative bubble in biotech companies.
Grade: Incomplete

"Biotech is really an ideal sector for mass speculation. It impacts everybody, and it’s inside everyone and everything: the TAM of “biology” is infinite. It’s incredibly complex, so it’s easy for novices to grasp onto a story and its importance without understanding the reality of it. And it’s full of high-leverage potential: it’s totally plausible that biological breakthroughs in health, manufacturing, fuel, etc could generate tens or even hundreds of billions of dollars of value.”

OK so, technically there was a speculative bubble in biotech right after I wrote this:

But, like, that doesn’t count and we all know it. That was an everything bubble; biotech just went along for the ride.

However, I absolutely stand by this call some time in the next five years, or maybe ten, because of this guy:

In a couple years, I think we’ll have made enough progress on AI applied to scientific discovery that people’s expectations around “what is possible in life sciences” to reset pretty dramatically. The less comprehensible it is to retail investors, the better. I’m giving myself an “Incomplete” here because it is actually too early to tell for this one.

#7: The only tech giant I’d be seriously worried about is Google.
Grade: B+

“Act One of Google was building the world’s greatest search engine, with the world’s greatest business model. Act Two of Google was colonizing the infrastructure of the internet, like a technological British Empire, in order to protect and fuel its crown jewel capital. The 2020s will be Act Three. As the decade begins, Google may well face the biggest real opponent they’ve ever fought – the United States – at a moment where its internal morale and purpose are crumbling and adrift. They print money, but if you ask me, they’re at risk.”

Welp,

I think I got this one more or less right. At the time I was writing this, everyone was really negative on Amazon and Facebook, who’ve since done fabulously. Google, on the other hand, had a vaguely uneasy feeling about it but was clearly in everyone’s conventionally good graces. Sure enough, the worries I expressed in that prediction have totally played out, culminating in the DOJ calling for Google to forcibly divest Chrome this past November.

The line I regret the most in this piece was “They engage in by far the most actually anticompetitive behaviour of any tech giant”. In hindsight, I wish I’d written instead, “Aside from Apple, who is far worse, but gets away with it.” I’ve learned a lot in the past five years about how critical Google’s work supporting Chrome has been to keeping the web healthy, and in a better state than the dark days of the browser wars.

The thing is, they’re still… doing fine? The stock is up 150% since I wrote this (along with its peers), so it’s hard to argue that the company is “in trouble” by that standard. Moreover, a little paper out of Google called Attention is All You Need went on to redefine the entire field of software and technology, and you hear a lot fewer voices today saying Google wastes their entire R&D budget. But maybe that underscores the point further: Google helped break open what’s now possible with LLMs, and Gemini is quietly pretty good. But it’s clear that Google isn’t disproportionately enjoying the outcomes here: all of its peers have gained as much or more in strategic options and value.

Overall I’m going to give myself a good grade on this even though Google clearly isn’t going to go out of business or anything like it - the trend was correct.

#8: There will be a significant moral panic around microplastics – not due to their environmental effects, but for their health effects.
Grade: A

“At some point in the 2020s we’re going to have another Leaded Gasoline crisis. We’re going to learn that some sort of chemical or microscopic thing that’s inside everything, all around us, actually has some horrible health effect we never knew about before. My pick for what it’ll be? Microplastics – particularly the microplastics inside clothing, like athleisure. The funny thing is, we already know that all of the fancy microplastic stitching and fabric material we put inside athleisure are terrible for the environment. That stretchy, comfy feeling comes at at an affordable price point, but a terrible environmental one – we already know this. But no one cares. We’ll start caring, though, when it turns out that they’re hazardous to us in some way we never understood before.”

Two parts to this one, and I’m gonna go ahead and say I aced them both. First all, there’s the moral panic component to it - which isn’t about ethics exactly, it just means, “This is going to become a popular cause”, and I should’ve originally articulated it that way, but that’s what I meant and I think it reads pretty clearly. Some people who’ve been ahead of the curve on this, like Nat Friedman, have already made a big impact on elites thinking of this as a problem. But as far as popular awareness is concerned, there’s no more obvious sign of “Things we are eating are poisoning us” as a popular meme than our friend here:

Moreover, I’m also going to claim credit for the prediction that the actual source of most microplastics in food - as far as we understand it - is actually from our clothes. For all our focus on processed food preparation, which is important for lots of reasons, the real source of this stuff is external.

Anyway, I’ll claim my first and only A grade for this one. I’ll need it for my GPA though, cause the next one was another clunker on my part:

#9: The 2024 and 2028 elections will be overwhelmed by “Campaign Hacking”: not by foreign governments or enemies, but by opportunistic hustlers and entrepreneurs.

Grade: D-

“We really haven’t yet seen the real extent of what the internet will do to political campaigning. I’m betting that the next creative wave isn’t going to come from the campaigners, but instead from opportunistic hustlers on the outside who figure out how to hijack them. … Hustlers and entrepreneurs, even more than politicians themselves, will figure out the formulas for how to force the media to cover you as a campaign story, almost certainly to the detriment of the primary process itself.”

Nope! Got that one totally wrong, at least so far. This did not remotely happen, as I’d imagined it. I will give myself a D as opposed to an F just because Trump himself has proved willing and able to keep doing all kinds of merchandising schemes as a part of his campaigning and executive style. But I had imagined a situation where the political establishment would actually lose some amount of control over what elections actually are to the reflexive power of social media and entrepreneurial hucksterism. That didn’t happen at all, so, bad grade for me.

#10: A real startup scam is going to seriously spook the Silicon Valley community.

Grade: A-

“I’ve long believed that the real mystery of Silicon Valley isn’t the outsider question, “How is Silicon Valley so wild and crazy”, but actually the insider question: “How is Silicon Valley so stable?” It’s built on speculative finance, it’s full of experiments whose outcome you can’t know for years, and it has to move fast enough and fluidly enough that (at early stage anyway) it effectively works on the honour system despite the FOMO environment. It’s so interesting how, in this environment, there aren’t any scams like this. … One of the important components of Silicon Valley working well is early stage investors willing to write checks without really having to worry about the downside – the worst you can lose is 1x your money. Financially, that’ll be true, scam or not. But reputationally? If a high profile scam becomes the talk of the town, there’s nothing scarier than everyone else knowing that you fell for it.”

I really struggled with what grade to give myself for this one. On the one hand, the prediction clearly came true, hilariously so, at a bigger scale than I could’ve possibly imagined.

That being said, I did describe something somewhat different from this in the prediction: a situation where investors fall in love with a genius founder(s), the founders illegitimately raise lots of money and abscond with it, and investors are left having to answer for “how did you fall for this?” And that is definitively not what happened with FTX. This wasn’t a fundraising scandal; unless you argue (correctly I guess, but pedantically) that FTT token was a fundraising instrument. You can absolutely argue that this prediction did not come perfectly to pass, on those grounds, and I’d accept that.

But the part of the prediction that did go right, in the funniest possible way, was the second-order consequences of “Why were you so captivated by this guy”. Which, in some ways, was the spookiest. It’s not just that everyone loved SBF, it’s that, 1) specifically, Sequoia Capital gave him the highest possible blessing you can get in tech, and 2) we had this amazing storyline of his genius transcending normal human rules. (There was that great story in the Michael Lewis book of SBF at Jane Street betting max-aggressive on every single round of the intern game, which everyone interpreted as “This guy is built different” but in respect may have suggested an alternate reading, “Maybe he’s bad at bet sizing?”)

But even funnier was the narrative of how good SBF was at League of Legends - how he’d be playing during VC pitch meetings, and the magic was working so powerfully that this was taken as a strong vote of confidence.

As it call came crashing down, the absolute funniest part was the revelation that his League of Legends rating wasn’t even that impressive, and anyone could’ve looked and seen that. The scriptwriters get extra points for making AOC deliver the crushing blow.

Nothing actually broke (as predicted), but the chill was very real, in exactly the way we’d thought: he became a pariah not only for how he ran his company, but for breaking a very sacred promise that Silicon Valley collectively makes to itself: that whatever happens, success or failure, we’re not going to spook the magic. Instead, now we’ll forever have the amazing artifact of baffled Matt Levine: “You’re just like, well, I’m in the Ponzi business and it’s pretty good.”

That’s it for now - I’ll check back in another five years for my final report card. How do you think I did? If you wildly disagree with any of these grades, I’d love to hear from you.

---

If you’ve made it this far: thanks for reading! I’m back writing here for a limited time, while on parental leave from Shopify. For email updates you can subscribe here on Substack, or find an archived copy on alexdanco.com.