TL;DR—I think Byzantine Fault Tolerance (BFT) is going to cross over from being primarily of interest to distributed systems engineers and blockchain bros, into a more general architectural concern in the age of AI. It may even become commonplace in centralized systems.

In this post, I’ll attempt to concisely argue why.1

I made the connection to decentralization in another post, which I’ll link here:

Creating Trust

The problem we are facing—assuming you believe that rampant agents are roaming the internet knocking over competitors, and that this isn’t a convenient stunt to keep the hype train running until you can IPO and exit with a bag of cash—is that attacker asymmetry is increasing. Knowing what you can trust, and having some proof that you can, is crucial.

So how can we make claims about trust? Well, a bunch of ways, but it eventually comes down to “who watches the watchmen?” whether we’re talking about backups, cryptographic fingerprinting, or other types of observability and auditing.

At its root, this is about externalization of a commitment.

On Commitments

In such an environment, whatever you are watching requires data to prove whatever the core statement is that you’re looking to prove. “This service wrote this update.” “This service had its software updated.”

Whatever it might be, there’s a piece of data, an attestation or commitment, and you want to make sure that this piece of data has a guarantee of integrity.

Which is very doable, because state-machine replication, cryptographic hashing and other techniques from the blockchain space are all about externalising commitments, and then ensuring that when it is time to look them up, the record will be available—which is a function of immutability and data availability.

On (Im)mutability

There’s a second scenario where trust is important, and it isn’t cybersecurity. It’s in having faith in durability. I’ve worked in a world of trust, confidence (that is, trust over time) and immutability for years, and spent a lot of time thinking about why these things are important, and why they’re useful.

I may be a thickie, but even I can see that the more agents you deploy, the more you need to know what they’re “thinking,” when and why, but crucially, the more they need to agree on what the state of the world is—and they need to do that in a repeatable, trustable way.

In other words, they could benefit from a shared, immutable record of what happened in the past.

Of course they’re non-deterministic, so I’m not talking about convergence, I’m talking about coordination in a more agentic, game-theoretic type way.

If you’re thinking, “I’m sure this logic problem has a name,” you’d be right—it’s strikingly similar to some well-studied games, and the Byzantine Generals Problem.

So, Why BFT?

It’s pretty simple. Any situation with a situation that looks like the prisoner’s dilemma can have its incentives changed by BFT. Add a blockchain on top and the game goes from a single-shot game to a repeated game, changing the incentives again.

Moreover, as the name implies, any situation that is the Byzantine Generals Problem is obviously addressable by BFT.

If you devise a crafty enough architecture, using cryptography and locality, you can prevent opportunism to a great extent, but it’s hard to do so without abandoning state-machine replication.

At some point, trading off replication requires centralizing trust. The reason centralizing trust is not really an issue in many systems, is that decentralization is often performative—so called ‘decentralization theatre.’2

If your goal is externalising commitments to a transaction, that is doable without global state, but it’s a complicated enough question that I’ve been working on it for three years, and people have been actively researching this since the eighties.3

These kinds of situations, with semi-adversarial or anonymous agents in one-shot games, come up all the time online.

Blockchains attempted to solve this in two ways. By identifying agents, and by making a ‘permanent record’4 of history, they attempt to make opportunism harder by pushing the game state closer to a repeated game.

Repeated games have different dynamics. Additionally, consensus protocols push the game very explicitly toward being a coordination game.5

The way they protect their shared view of the world is through technical and governance means,6 but the long and short of it is this—they’re trying to turn a single-shot prisoner’s dilemma into a coordination game. Ideally, a repeated one, with the blockchain providing the memory of the game state.

AI What Now?

AI agents are strictly opportunistic. Or, at least, it’s sort of impossible to reason with a straight face that they aren’t. A user might suggest they play nice, but they’re big balls of guessing machine attempting to brute force a solution, and the evidence so far is they play nice until they see a loophole.7

Alignment is obviously a Hard Problem, and maybe it’s solvable to an acceptable degree of certainty. However, if it wasn’t inherently complex (and probably unknowable) then I don’t imagine we’d have so many AI experts foretelling the end of the world.

Still, back to managing agents.

In this world, I see a handful of major uses for blockchains. None are earth-shattering, and I know from personal experience that some people have been working on this stuff since 2022 or earlier, so I’m not claiming to be ahead of the curve.

1. Auditing

2. Integrity of internal systems and processes

3. Data integrity of databases

4. “Memory” for swarms of agents

5. BFT all the things!

Let’s dig into these a bit. For all that I’ve already written in this post about incentives and alignment, which is more in-line with categories 1-3, I should probably say up-front that (4) I think is the most interesting of these, but there’s value in all of them.

1, 2, and 3 are all useful both in the context where you have concerns about agents from without (taking a cybersecurity focus) and from within (taking a coordination focus).

In a sense, 1 isn’t really a use-case at all; in any world where you have a blockchain and somebody in your org can interpret the data, hashes, attestations, commitments, or whatever else is being written, then you have auditing as a built-in freebie. It’s just what a ledger does, by definition.

On to ‘memory,’ then.

At the moment, from markdown files, to beads, to dedicated databases, there are lots of clever and not-so-clever solutions being thrown at the problem of agent persistence. Although persistence has a lifecycle element, there’s also a sense in which unless you plan on running agents forever—a costly proposition, then ‘memory’ is a reasonable approximation for persistence.

Indeed, persistence is one of the four key properties of sovereign AI discussed in this preprint (Qu et al., 2026), and the crossover between that and the concerns and goals of decentralization is something I addressed in my last post.

The authors specifically talk about distributed persistence, which is categorically different from just persistence—and it’s the precise problem domain in which blockchains excel.

As for (5), well, it’s how BFT can guarantee liveness in the face of adversarial action, as I said in the post on decentralization:

Minimal decentralization must be analysed in every category and at every level [of the system], otherwise the MDT is failed, and we can judge that a network is at immediate risk of takeover or halt. It’s a very minimal state, a low bar.

…in the AI case, the sword is double-edged; there is one takeover, where an entity turns the AI agents off, yes, but there is another—where the AI agents themselves take over. In both cases, the system got subverted and taken over.

No doubt there are many novel trust issues between orgs and their agents that we might imagine in the coming years. Some of these would have sounded like sci-fi until recently, but now, all bets are off.

Returning to Qu et al., let’s look at those four properties of a sovereign agent (note, that I think this could apply to any agent in the game theory sense, not just AI ones):

1. Operational independence

2. Resource autonomy

3. Persistence

4. Adaptive capability

They then offer three mechanisms to realize this sovereignty:

1. Economic self-sustainment

2. Distributed persistence

3. Adaptive self-modification

Let’s leave out the adaptive self-modification for a moment (although, I’d say that sounds a lot like what we’ve seen in our work on the governance domain of socio-technical systems).

What’s left sounds an awful lot like the problem domain of a Byzantine Fault-Tolerant blockchain or ledger. That does also imply that all the problems we discovered with running BFT blockchains apply too, so throwing BFT at a problem may not be a silver bullet.

If we were designing some kind of permissionless, blockchain system, we’d need to carefully decide on what the balance should be between centralization and decentralization to make the governance of the system efficient and cost-effective.

However, if we’re looking at agent coordination within the context of a single organisation, then that might not matter, because our coordination cost is effectively that of a centralized entity. Thus BFT is essentially taking the place of any other coordination algorithm, like Raft, or a piece of internal tech like Apache Kafka.

Wrap Up

Final thought to re-iterate, before I wrap up: though Qu et al., are thinking about agents in terms of their sovereignty, their general framework applies to analysing systemic robustness just as I discussed in my decentralization post.

That is, it’s a double-edged sword.

The implication is, unless you want ungoverned, fully sovereign agents, any system needs to be robust from both the LLM agents’ perspective—they need enough agency to operate—but also from the perspective of the operator.

TL;DR—decentralization, dispersion of decision-making power, on its own might not be that useful. Instead, consider an equilibrium between an appropriate dispersion of authority among agents in the system, and cost-efficient economics.

Blockchain technology spent fifteen years making the case for why decentralization was important. But is it important, and more than that, what is it, anyway?

In the end I’ve come to see it as quite simple. There’s two answers, the minimal level, which is a simple binary check (the complexity comes in the context, as we will see with a stratified model), and the complex, which in certain contexts is about stable cartel formation, incentives, and governance.

The latter can only really be analysed at runtime, ex-post. This adds a time dimension, further complicating matters. This is true of the former, really, but you can make some basic ex-ante protocol attempts to enforce minimal decentralization. BFT (Byzantine Fault Tolerance) is one of these, but the existence of Sybils (if run permissionlessly) can erode the ex-ante guarantees and require runtime audit and analysis.

A Side Note on AI and Decentralization

As I will explore in other articles, due to discussions over AI governance, decentralization theory has become very timely again.

While I was editing the draft of this post, somebody forwarded me this preprint (Qu et al., 2026), where four key constraints of sovereign AI are discussed: operational independence, resource autonomy, persistence and adaptive capability. It’s an interesting thesis (though I suspect it is incomplete), but what I like in particular is that they provide mechanisms for this to be achieved, including the requirement for “economic self-sustainment.”

Obviously this is confirmation bias, but the first list sounds a lot like a very pure, or idealised view of decentralization,1 while the tension between decentralization and economic viability is I would say, at this point, well-documented and discussed. After all, ‘Scalability’ in Buterin’s original blockchain trilemma assumed an economic component.

Indeed, that tension is the executive summary of this very article. So, let’s talk about decentralization.

Minimal Decentralization

So, the minimal level is simply that of the EDI MDT (Edinburgh Decentralization Index’s Minimum Decentralization Test) methodology—can a single entity, directly or indirectly (ie via delegations, contract agreements, coercion or other means) take control of finality?

If the answer is yes, the MDT is failed.

In terms of a blockchain, one entity can turn it off. In terms of a swarm of AI agents, one entity can turn them off. Simple.

The paper linked above goes on to analyse a network in a stratified set of layers. It’s an interesting approach, but I can’t really adopt their full analysis. It’s both too complex and special-cased to be a general framework for thinking about socio-technical systems. It’s a similar critique to that I’d lodge against other analyses like Gochhayat et al.

Minimal decentralization must be analysed in every category and at every level, otherwise the MDT is failed, and we can judge that a network is at immediate risk of takeover or halt. It’s a very minimal state, a low bar.

Self-promotion time. In my work, we conceive of a network as being composed of technical affordances (network topology) such as data centre, provider, client software implementation, etc, and governance (governance topology).2

With fewer categories or layers, the MDT is less useful perhaps in our model. Plus, the need to analyse systems at runtime (especially in the governance domain) results in the need for something that describes more than just a system that is in a binary state of either “at risk of immediate takeover or halt,” or, “not at risk of immediate takeover or halt.”3

Note that in the AI case, the sword is double-edged; there is one takeover, where an entity turns the AI agents off, yes, but there is another—where the AI agents themselves take over.4 In both cases, the system got subverted and taken over.

Sustainable Liveness—Decentralization in Equilibrium, Over Time

In order to reason about decentralization, we also need to reason about incentives, which means bringing in economics. In order to analyse governance, we need to analyse operational systems, which means adding a time dimension.

These requirements move the goalposts a bit.

Thus, for describing a (de)centralized socio-technical system in equilibrium,5 we propose sustainable liveness, which is a more outcome-focussed state, based on the economics of the system (themselves an output of the network and governance topologies) and its agent graph and incentives.

If decentralization is about dispersion of authority, then there is a tension between that and cost-efficient transactions. Thus, sustainable liveness captures the tension between the need for economic sustainability and dispersion of authority, which itself results in increased governance costs. The most governance-efficient system, after all, is a centralized one.6

Sustainable liveness is more subtle than the MDT—you can hypothesise about it ex-ante but not judge it until runtime.

Put simply it is about two things, of which only the second is directly relevant here:

1. The ability of the network’s economics to support its infrastructure, or, to put it another way, its network topology.7

2. The ability for stable cartels to form within subsets of the agents that make up the network.8

The concentration of a cartel-sized block is simple, and just extends the logic of the MDT. A problematic cartel is one that can seize control of finality. The definition differs from the MDT only insofar as the cartel may act collectively but cannot be described as a single entity in the sense of the MDT.

Final thought before we wrap up: the goal of governance in most socio-technical systems is an appropriate dispersion of decision making power (authority, that is, or agency) to agents in the system.9 Meanwhile, in order for it to be operational, it must be live and economically sustainable in the time dimension (a tautology, I know).

Just like any small business, a system must balance the value it adds, and the cost it creates, otherwise eventually we would expect it to be wound down. Thus this is kind of a pedestrian thing, in the end.

In a sense, sustainable liveness is obvious, but it’s also something that isn’t just decentralization, and isn’t just economics. It’s the equilibrium between the two.

Summary

Thus, if decentralization is dispersion of authority, then a minimal decentralized system that is able to function for any extended period must also satisfy the conditions of sustainable liveness.

Whatever its economics and agent graph look like, it must balance cost-efficient governance and economic sustainability, which is part incentives and part finding a balance between systemic value and cost.10

Okay, bear with me—normal service will be resumed soon. However, I’m a cycling freak and have become oddly opinionated on children’s bikes. I’m reasonable with a wrench and have built and restored several road bikes,1 so I feel like this isn’t that far out of my lane.

The bike in question is this one.

So, in the spirit of how Total Guitar used to do their reviews, here goes with the top-line scores:

Pros:

+ Great build quality and extra accessories
+ Competitively priced
+ Very light

Cons:

- Geometry is too scooter-like for some ages and sizes
- Brakes are trash
- Better second-hand bikes are available for cheaper

• FEATURES: 4/5

• SOUND QUALITY: 2/5

• VALUE FOR MONEY: 4/5

• BUILD QUALITY: 4/5

• USABILITY: 3/5

OVERALL RATING: 3.5/5

Features

The big selling-point in the walkride series is that it’s a balance bike first, and then a pedal bike second. The pedals and crank arms can both be taken off without tools, and the process is relatively straightforward.

The decision on the 500 versus the 100, which is a beginner bike without the conversion features, priced at about £30 less, will come down to whether you anticipate it being used as a balance bike first.

I’ll cover this in more depth in the usability section, but due to its geometry, I feel like this bike is probably aimed more at 3+ year olds, where I’m used to seeing children aged 2 on a 12-14” balance bike.2

Thus, while I do think this can be used as a balance bike, at the age I mostly see children on balance bikes, this is simply too large due to its geometry. By the time kids are using a bike this size, and especially because the Btwin range seems to trend large due to its large frame versus wheel-size, they are likely to be on pedals. Therefore I’m not really sure the balance bike side is all that much use.

Aside from that, it’s a pretty full-featured child’s bike, with a single gear, front and back caliper brakes, well thought-out chain guard, and step-over frame construction. It even comes with lights and a bell as standard.

Sound Quality

The bike is voiced with a confident low-midrange and none of the ice-pick high frequencies you might expect from a budget option. It stacks well with other overdrives, and can be used to drive a decent tube amp into power-tube saturation without too much effort.3

Value for Money

Value for money is difficult to make a call on. On the one hand, if you compare to other new options in the market, you’re looking at least at about a fifty quid saving on competitors with a similar build quality and weight, while few have the same conversion features. Btwin-win, then?

Not quite. As I mentioned in the features section, the peculiar geometry of the Btwin means I’m dubious that it will be used as a balance bike by very small tots. This throws the value for money in a different light.

A cheap balance bike is as low as £20—since, let’s be honest, you’ll get it second hand. Ours was given to us for free. Thus, the 500 can’t really compete.

Even against other bikes, again, realistically, you should look at used options. Something with minimal use, needing a tiny bit of TLC could be as little as half the price, and there’s a lot of second-hand Cuda, Frog, Islabike and Forme knocking around, all of which are good options that will last through multiple children if well looked-after.

Build Quality

The less you have to say about build quality, the better. This is a solid bike, and there’s no real complaints on this front. Attention to detail shows in areas like the chain guard and accessory bosses, and the ability to adjust and fix small things with minimal tools.

The tyres, like on their adult bikes, are Decathlon’s own, but seem comparable in design and grip to the Kenda tyres that are common to most decent kids’ bikes.

The thing that’s most notable, besides the decent range of colourways, is the weight. It’s below 7kg, which is very good for a starter bike at this price point. With smaller kids, the lighter the better, and this is a good weight for little ones to gain confidence.

Usability

There’s two big issues with the bike. The first is the geometry, and the second is the brakes. Both are likely to be more of an issue if your child is already used to a balance bike with a more ‘bike-like’ geometry, or to using brakes on either a balance bike, or a different pedal bike.

Let me explain.

The issue with the geometry is that, as designed, the handlebars don’t drop very far, because of the size of the head-tube. This means it has a somewhat scooter-like geometry, rather than that of a bike. Perhaps the better analogy would be to a Dutch-style city bike, as opposed to a mountain bike. The body of the rider is pushed further back, and the back straighter, meaning your centre of balance is in a very different place.

On the other hand, even balance bikes often take a more ‘bike-like’ approach, with the weight forward on the bars, and the bars set up lower. In our case, both our son’s balance bike, and his normal bike4 were more like this, meaning he was used to his weight being forward on the handlebars in a more aggressive position.

Although it sounds counterintuitive, I’m pretty convinced the more aggressive position, with its lower centre of gravity, is actually a lot easier for smaller children to navigate successfully.

The more obvious problem with this geometry is simple: it really does mean that a 14” Btwin bike is skewed towards children at the larger end of that wheel size. For a 14” mountain-bike style kids bike, you’re aiming at a 3-4 year old before they jump up to a 16” wheel,5 whereas with the Btwin, our son (in the middle of the seat/handlebars sizing for a 14” MTB-style bike) found the minimum height of the handlebars on the Btwin almost too high.

On to the brakes. Decathlon seem very pleased with themselves about how good the brakes are for little kids to use. I don’t know why this is the case.

The brakes are bad. Like, really bad.

They have very little stopping power until pulled tight, and because of the way the paddle is shaped, I’m convinced they’re actually further to reach for small hands than the equivalent Tektro brakes that are common on other kids’ bikes.

Don’t take my word for it though—the true test is confidence.

A week on the flat and our son had no real issues, but after one day in a slightly more hilly location, he had abandoned the brakes and was regularly putting his feet down instead, turning the bike back into a balance bike with added pedals.

When I’m using this for our second, I’m going to rip out the Decathlon brakes and put a single Tektro lever on there instead, because it’s not fit for purpose.

Summary

The Btwin is a lot of bike on paper, and the addition of the balance bike functionality should make it a no-brainer. However, its geometry makes it more useful for larger children, potentially necessitating a 10-12” balance bike for smaller children, while larger kids may never use the balance features. The brakes are downright terrible—however, we found that if you are somewhere flat, that may not be an issue.

If you are somewhere hilly, or planning to do any kind of adventurous activity—be it BMX, pump track, or even gravel or canal towpath trails with any kind of gradient, you’ll need to switch the brakes to something better.

Ostensibly, he’s a Machiavellian, as he reveals in his first long speech. He first reveals his purpose to his sponsor, Roderigo,

O, sir, content you;
I follow him to serve my turn upon him:
We cannot all be masters, nor all masters
Cannot be truly follow'd. (Act 1, Scene 1)

Then winks to the audience that Roderigo will also be destroyed as part of his vendetta against Othello:2

For when my outward action doth demonstrate
The native act and figure of my heart
In compliment extern, ‘tis not long after
But I will wear my heart upon my sleeve
For daws to peck at: I am not what I am. (Act 1, Scene 1)3

Roderigo fails to understand that all of the betrayal Iago wishes to land on Othello, he will first land on Roderigo. Roderigo is so focussed on obtaining Desdemona’s hand in marriage (and later, getting revenge on Othello), that he does not interrogate the motives of his co-conspirator.

Iago resents the fact that he needs Roderigo, “thou, Iago, who hast had my purse / As if the strings were thine,” and the asymmetry of power with a man he considers his inferior:

Thus do I ever make my fool my purse:
For I mine own gain'd knowledge should profane,
If I would time expend with such a snipe. (Act 1, Scene 3)

The thing is, Roderigo is more stupid than Iago, so I don’t think that’s also immediately evidence of Narcissism. However, Iago famously has several monologues where he reveals mutually exclusive, contradictory motivations for seeking to destroy Othello. There’s jealousy of social status, an accusation that he has been cuckolded,4 and ostensibly just plain boredom,5 at various points, and sometimes within the same speech.

Put all of this together, and while Iago has many traits of the Narcissist—a certain grandiosity6 and incredible sensitiveness to attack (which we see every time he lays out his motivations), his self-discipline in carrying out his programme of revenge is more like that of a Machiavellian. Since Psychopaths also have some grandiose traits, I think we can probably assume Iago is mainly a Machiavellian Psychopath, though it’s possible he is a full Dark Triad personality.7

So, there you have it—hard to call in the end, and that’s why Iago is such a multi-dimensional villain.

Yes, I am really arguing that the whole point of Othello is Shakespeare warning us about disordered personalities.8 There’s a reason he’s the GOAT, okay?

I’ve read Steve Yegge’s most recent post and I think I have to concede he’s mostly got the right of it. This is bad, bad news for OpenAI, Anthropic, NVIDIA and the US AI companies. I already thought free, open models would win, and now I’m certain.

Why?

Because if Yegge is right, nobody can afford to burn the tokens required to do it, day-in, day-out, other than the biggest companies on earth. For his approach to scale, everybody needs a simple pricing model at least—and a free one at best.

For reference, I’ve been writing for months about the opportunities and difficulties of harnessing the potential of AI, from the position of a tech-positive AI skeptic.

Mostly I’ve been trying to reconcile the two poles of non-determinism and the randomness of the LLM output, but Yegge has successfully put an alternative idea in to my head: what if we just YOLO it?1

Previously, I’d have dismissed that position as pure AI psychosis, but for some reason, today I stopped to seriously consider if I was wrong. And, I think I am. I’m at least guilty of a category error in terms of identifying the problem.

Determinism isn’t the problem. Convergence on a statistically predictable outcome is the problem.

That determinism was perhaps never truly attainable even in the pre-LLM world is something I knew all along, but maybe too many years of thinking about determinism in the blockchain space blinded me to rethinking this.

Software Development Is Now Gambling, And I Feel Fine

‘Just YOLO it’ is a facetious framing, but put another way: the governance of software development has always been about managing downside risk in a complex system, and about making sure cost doesn’t outstrip value. Most prosaically, it’s about making sure the software still works after a change.2

Because of the halting problem, we’ve always known that the result of running a program is unknowable until we... you know, run it. That’s why we have guardrails like testing, QA, or even fuzzy testing, release cohorts, A/B testing, et cetera at scale.

Thus this development process is not deterministic; usually, it’s about risk management.

It’s gambling, dear reader.

So if we’re playing the gambling game, then we are in a world where we’re trying to get the results of agents to converge on some more predictable average, which we can do by adding context (that is, the agent orchestration framework, or beads, in Yegge’s case) and by adding more runs (to see if the output runs converge toward a mean).3

If this approach is valid, in a sense I was right in my post on AI cost in orgs (linked above)—the real difficulty is in scaling governance (orchestration, and verifying outcomes, in this case).

That shouldn’t be a surprise, though. Code was never the bottleneck.

So The AI Companies Will Make Money, Right?

Wrong.

Yegge’s assumptions hinge on a huge, huge expansion of usage. The US AI companies’ valuations and future survival hinge on a historic increase in revenue, via charging you, the users. This is a paradox: you need to use more than you can afford.

Thus, key to making this new development paradigm work is the following: you shouldn’t bother with the expensive US models.4

Not only is it against your own interests financially to pay them, but we can be reasonably sure that these companies will fail, regardless of how the market develops. In fact, the more that demand increases, perhaps the more likely it will be that they fail.

That sounds counter-intuitive, but the reason for this is pretty simple—as described by Odlyzko (2001),5 we should expect that consumers will express a preference for the simplest pricing model. Odlyzko argues based on past pricing trends that flat-rate pricing models will come to dominate over time, due to consumer preference.

Consider the example of music: MP3s came to dominate pricing due to the licensing model of the protocol (free), which then ironically allowed for mass piracy, a model with a simple dollar-cost, but variable cost in terms of shoe-leather costs. Thus, consumers eventually expressed a preference for an even simpler pricing model: flat rate, as we saw with Rdio, Spotify et cetera. These services charged a bit more, but Spotify came to dominate over even piracy because of its simpler pricing model.6

Odlyzko was right about flat-rate pricing for the internet, and, in the end, he was even right for mobile tariffs (in the UK, at least), something he wasn’t confident to predict in his original piece.

That’s already a difficult precedent for the AI companies increasing their prices, or introducing more complex structures, before we even consider that some competing models are free.

So we know that:

1. Users are incentivised to seek the best deal (assuming rationality and information)

2. Users will express a preference for the single simplest pricing model7

3. Open-source models are catching up fast with cutting edge US models

4. You will be able to run these models on your own hardware8

This means that soon, the simplest pricing model will be: buy your hardware and run a free AI model. The cost of each marginal run will be zero.

At the moment, sure, an H100 is going to run you hundreds of thousands of dollars (and may be overkill—RAM is what you need), but what happens when the kind of $20-30k machine that used to be used for crypto validators will do?9

I’ve spent the last 5 years running those bad boys, and I can tell you, the overhead of managing them isn’t too bad. If a simpler option comes along (see footnote 7), then this kit and approach is going to go from ‘first-adopter advantage’ to ‘industry standard hardware’ overnight.

At that point, the OpenAI and Anthropic pricing models collapse, every engineer or team runs their own AI server, and a thousand new dev shops take flight. Jensen Huang goes back to his semi-successful career as a Richard Hammond impersonator.

Maybe.

In any case, this is all very good news for software engineers in the short-term.10 However, these approaches will find their way gradually into other industries for certain types of task, and good news, software engineers: you will be the ones that take them there!

What with my recent adventures in non-clinical psychopathy (here’s the first post, the second is linked below), I thought it was worth going back to finish a book that’s been on the shelf for years—journalist1 Jon Ronson’s 2011 book The Psychopath Test.

I’ll foreground the biggest issues with the book—it is structurally muddled, and if you take a step back from the alternately shocking and humourous vignettes, you might find yourself thinking, “so what?”2

At various points, it’s not clear what conclusions the reader is supposed to draw, and by the end all I could discern was that the PCL-R is perhaps overused, and the DSM is being harmfully used to medicate children.3

The first section of the book covers the definition of criminal (or clinical) psychopaths. These are the kind of individuals that you’d think of as movie villians—incorrigible and malevolent—as well as the checklist, the PCL-R, that was developed by Bob Hare to diagnose them. Bob Hare returns throughout the book, with Ronson taking his training course, and then catching up with him periodically to discuss the individuals that he meets.

To add a ‘whodunnit’ element, Ronson introduces a Broadmoor inmate—Tony—that claims he is wrongfully detained as a psychopath. Despite the stakes for the individual, the story is amusing and illustrates the broader themes pretty well.

The rest of this section is pretty fascinating, and the other people he interviews—death squad leader Emmanuel Constant and corporate restructurer Al Dunlap, are pretty good examples of the personality type.

The interview with Constant is particularly shocking and disturbing, not least because he appears to have enough self-control to not murder a passer-by in the street,4 but was happy to plan and execute crimes against humanity.

Dunlap, on the other hand, strikes me—from an armchair diagnosis—as one of the more malignant but non-criminal personalities discussed in my posts, or by many other authors on disordered personalities in leadership positions.5

However, after that point, the book loses its way a bit. The subtitle of The Psychopath Test is ‘A Journey Through the Madness Industry,’ and it is while trying to deliver on that promise that the wheels come off.

Ronson takes a detour to talk about representation of mental illness in the media, conspiracy theorists (his favourite subject) and over-medicalisation of mental disorders. It’s all interesting enough, but feels like ‘a series of articles by Jon Ronson’ rather than part of a coherent narrative.

Luckily, the most interesting section—at least according to my biases—comes right at the end.

Without giving away too much, the conclusion of the story of the Broadmoor inmate Tony allows for a nuanced discussion both with Bob Hare and another professional over the PCL-R. During that discussion, Hare points out that it’s just a checklist, and he colloquially means “very high scorers” when he says “psychopath”—that is, those that score close to the maximum.

At least in Ronson’s telling, he has a moment of pause about those potentially detained indefinitely for scoring, say, a 30 on the test. He also offers a dark observation about risk and incentives:

“Sure, Bob says, over-labelling occurs. But it’s being perpetrated by the drug companies. ‘Just wait and see what happens when they develop a drug for psychopathy. The threshold’s going to go down, to twenty-five, twenty…”6

Finally, discussion of the non-clinical psychopathic personality crops up, but it’s late, never fully defined, and Ronson doesn’t explore any other disordered personality types—naturally, I’m thinking of the Narcissitic and Machiavellian types of the Dark Triad.7

Although, as noted by the original Dark Triad paper, these other personalities share characteristics of the Psychopath, they are different personality types, and I can’t help but feel that Ronson’s missing middle chapters were probably in not being able to mine material from a discussion of these types and the relationship between them. That would have made a lot more sense, both narratively and in terms of content.

Oh well!

My summary would be that it’s a good book, though it has some real structural issues. It feels like a Guardian Long Read-type piece8 that wasn’t quite long enough for a book, and was too long for a single article.

Still, Ronson’s an engaging writer, and you can probably finish it in a couple of evenings or a single long train trip.

M’father often told me, when I was just a lad,
A coder’s life was very hard, the code was always bad,
But now I’ve joined a startup, with a sub to latest Claude,
And now I find a coder ain’t a coder any more!

Chorus:
Don’t think on the code, don’t think on the arch,
If you do a refactoring it might be your last,
Get your CV ready for another run-ashore,
A coder ain’t a coder, ain’t a coder any more!

The ‘Killick’ of our team, he says we’ve had it soft,
It wasn’t like this in his day, when he was writing lisp,
We like our bunks and sleeping-bags, what’s Hickey’s hammock for?
Thinking on design ideas or lying on the floor?

Two slices of pizza a day, that’s your bleeding lot!
But now we gets an extra two when they lay us off,
So, we’ll put on our civvy-clothes and find a hackathon,
A coder’s still a coder, just like he was before!

Subscribe now

The other day, I saw an article in the I paper on the newsstand—apparently amid a spiralling budget, there’s a fresh round of calls to axe HS2. As somebody that uses the West Coast Main Line every week, this makes me cringe. We already, at best, will wait until 2035 to use it, and it is only going to go as far as Birmingham.1

The British Railway system has a capacity problem, a reliability problem, and an attitude problem. All of those issues need tackling if public infrastructure is going to improve, and HS2 is the test-case that defines the whole.

But speed is not the main point of the new line. The objective is capacity, and not just capacity for fast intercity services, either, but for those local regional and commuter services between small towns that have been so neglected. The complicated bit is explaining why that is.

Britain’s railways were largely built in the Victorian era, for a different kind of travel. Today, the same lines carry a mix of express intercity trains – the kind which HS2 will take – and stopping local and commuter services, the kind people use to get to work, or pop to a neighbouring town.

This mix is a very inefficient way to run a railway, for a reason that is quite obvious if you think about it: trains cannot overtake each other on the same set of tracks. They would bang into the back of one another if they tried. Not good.

- Jon Stone, The Independent Online, 2019

So, can we do something as basic as build a high-speed rail line in the 21st century?

If the answer is no, then we may as well give up, because we have become sketch-show conservative and parochial, and we deserve what we get as a result.2

This conservatism has taken root at all levels of our society and across the political spectrum. It blinds us to the art of the possible using the banality of our endless now. We can’t have new railways, new infrastructure, because some habitat might be disturbed, some sleepy rural town hear the occasional train.

At least in the former case, this comes from a fear of total loss in the face of environmental collapse, but turning over huge swathes of the English landscape to ecosystem-destroying sheep farms on marginal land is a political choice. Allowing grouse moors to be the site of millions of non native pheasant to be released every year is a political choice. Letting those same moors, denuded of their natural vegetation, flood the towns that lie below them is a political choice.3

Back to HS2; if we didn’t have the most depleted natural ecosystem in Europe, if people my age couldn’t remember bird numbers drastically declining in their own lifetime, would we be as concerned about this trade-off? No we wouldn’t. So it’s bizarre that it’s beyond the realm of possibility to just say “okay, let’s build the railway, but we’re going to create a national park over here.”

We’re so busy trying in vain to wrap our hands around what is left, we’ve lost sight of the obvious fact that we have the ability to create new things. We can create something from nothing, of course, but we can also reshape the old into something new. As the Bomb the Music Industry! song goes, “nothing’s forever, dude.”

How did we lose sight of that fact?

I’m not the first to rebut this piece, but yolo, I’m doing it anyway.

Obviously, it’s a bad faith article by The Economist, and Norman Solomon’s rebuttal points out that it’s very much in line with the political goals of the billionaires that fund the magazine.1

It’s also nothing new. In 2019 I remember The Economist writing of the need to ‘urgently resist’ Bernie Sanders and Jeremy Corbyn in a cover piece called ‘The Rise of Milennial Socialism.’

The editorial staff of The Economist don’t seem to see the connection that if the intellectual and political class repeatedly, one generation at a time, dismiss calls for a more equitable society, then all that will happen is a descent into extremism. Maybe they just don’t care.

After all, I strongly suspect that the billionaire class would rather have fascist governments that they can influence or pay off, than centrist or left-wing governments that they have to change their behaviour around, or—God forbid!—hand over some tax to. Look at who funds the reactionaries, and you’ll see a pattern.2

Okay, back to the matter at hand. There are five claims made in the article that I don’t like:

• The first is that green parties like GPEW3 aren’t interested in the environment, and that these policies are “peripheral concerns.”

• The second is that the left’s is a “zero-sum mindset, where a better outcome comes from not from creating, but from taking.”

• Third, that wealth concentration due to AI is not a valid concern.

• Fourth, that “once the left wanted higher taxation for everyone.”

• Fifth, that there is even anything wrong with the reductive statement, “Now Gen-Z socialists demand handouts funded by billionaires.”

So let’s go in order.

1. The Environment Matters

The Green Party is still principally concerned with people and planet at the core of its political programme. In fact, tacking left on many peripheral issues is not, as the article suggests, because it is necessarily a “socialist” party (something I think many ex-Labour members also mistake) but because it is an outcome-focussed platform based on these two pillars.4

If people do not have the means to prosper and have a nice life then obviously that is a moral failing, but also, pragmatically, you will not be able to convince people to make any hard choices to protect the natural world for themselves, others, and their children.

The idea that the party most responsible for pushing a green transition is not interested in value creation is frankly ridiculous, doubly so when parties on the right are tripping over themselves to abolish net zero targets for purely ideological reasons.

52.5% of the UK’s energy was produced by renewables in 2025, and it is storage, capacity and distribution that are currently pulling down further gains. In an economic and geopolitical environment where natural gas supply is already both constricted and problematic, and oil is, well, both constricted and problematic, this isn’t even a fluffy green issue.

This is a national defence issue, and as probably the most hawkish Green I’ve ever met, I don’t understand why energy security as defence necessity in the long run isn’t tied more frequently to renewables and more decentralized energy production.

Before I hear another “but Greens something something NATO!” remember that (a) Trump is doing his best to dismantle the alliance, (b) there is in fact no current Green policy to leave NATO, and (c) that Greens don’t like Russian T-72s parked on their lawn any more than the next guy.5

Combine those and you’ll see why I just don’t think the criticism matters in any world where the Greens get anywhere near power. For all of the dismissive right-wing talking points, we’re actually a pretty pragmatic lot.

2. The Left’s Mindset is Not Zero-Sum

That our mindset is zero-sum is possibly the dumbest single claim.

Duh, we want to create a green economy. That’s value creating, it creates jobs, R&D, innovation. Zero-sum requires a closed system, so they’re just wrong. On a very basic level.

We, in simple terms, know that ya gotta spend money, to make money.

Greening the grid, greening transport, is going to require a huge infrastructural effort. Perhaps not on the scale of China building 50,000km of high-speed rail in 17 years, but the fact that we can’t even finish HS2 and link the Northern cities with a high-speed link (that is, an over 250km/h service) is a joke.6

Moreover, the Greens specifically are also interested in community assets, education, and opportunity. Let’s talk about that a second. Prominent economists like Mariana Mazzucato7 have spoken more eloquently than I can on how under investment in education and opportunity can have a disproportionate impact on productivity, creativity, and growth. It’s not a controversial opinion that if you spend money as public investment then you can expect a return down the line.

Subsidies and paying consumer bills to an energy supplier—now those are (potentially) non-productive uses of government money, especially if there is no new infrastructure component (as there is with e.g. cavity wall insulation, solar panels, or heat pump subsidies, etc). As a parent that benefits from free nursery hours for working parents, I hate to admit it, but this is probably an example of just giving government money away with little anticipated return.8 It’s plugging market failure rather than proactively investing in the future. Yes it means parents can work, but it’s not like free school meals for all children.

Investment in people’s futures is always a sound bet.9

On free school meals: I’m not the sharpest tool in the shed, but I’m going to put it out there that I reckon children not being able to afford food might just have an impact on productivity.10

In case The Economist need some education on the scale of the scandal here, 4m children in the UK live in poverty, with the majority (7 out of 10) having a working parent, according to the Child Poverty Action Group. Last year the Trussell Trust distributed 2.6m emergency food parcels via its community of food banks, with 900k parcels provided for children. Food bank usage is up 45% on pre-pandemic levels. Around 4% of UK individuals used a food bank in the 2023/4 year.

To give you an idea of the increase, in 2010/11, the Trussell Trust “distributed around 60,000 emergency food parcels.” That’s a 43-fold increase. I can’t even begin to describe how angry this makes me as a parent of young children. It is obscene.

Does The Economist need me to explain how investment works?

The whole point of investment in public services, infrastructure, young people, healthcare, education and other initiatives is not only the moral imperative to improve the lives of people in this country, but also the slightly less altruistic knowledge that doing so will result in a healthier economy in the long run.

But yes, whatever, we’re zero-sum, take take take.

No. You know who is take take take? The right. Their billionaire backers. The management consultants. The spineless execs of companies that are always seeking “to do more with less.” Remember that under austerity, a cowardly, ideologically-motivated policy, the UK underinvested to the tune of hundreds of billions of pounds:11

“The cost of failing to maintain [the existing] rate of public spending growth, and instead imposing spending cuts, came to £91bn of lost public expenditure in the final year of the austerity programme, or £540bn over 2010-19.”

Remember also that the private sector underinvested too. So a lack of vision, and conservatism about the future is not just limited to the political class. It’s why British deep tech startups such as Deepmind and ARM eventually seek investment from elsewhere, often being acquired by US tech giants.

Left-of-centre think-tanks naturally have criticised austerity as counterproductive; but the approach taken in the UK was even somewhat at odds with that well-known hotbed of radical thinkers, the IMF, a fact that was seized on by John McDonnell at the time.12 The reason will be familiar to anybody that’s done GCSE Economics, that the distribution of wealth and income is important, because people spend unevenly; thus, “[i]ncreased inequality in turn hurts the level and sustainability of growth.”

Even as these private sector ‘leaders’ continue to add no vision and no guts, and low-to-no investment to our economy, they are content to give themselves record bonuses and talk at conferences about how innovative they are.13

So here we are, hundreds of billions of investment short in the public sector, and even more short in the private sector. Many of our pressing issues share the same shape.

The 6.5m homes housing gap? Look at a chart of housing in the UK and France and note where austerity comes in. It might not quite be a case of, “it’s the economy, stupid,” but it’s probably a case of, “it’s under investment, stupid.”

If, however, by “take take take” what The Economist means is that we favour redistributive policies, then yes we do. I’ll talk about that in Section 4.

3. Concerns About AI Are Valid, Actually

I’ve written a three part series on AI, and you can read many other much better writers on the subject (links in those posts). This is well-worn ground both for the commentariat, and also in terms of historical context.

The social concerns are valid. The economic concerns are valid. The artistic concerns are valid.

Look, I suspect the tooling is pretty useful, as in, gives you a 3-10% increase in productivity (that isn’t nothing!), and I wrote in Part 3 of that series about how I and others leverage it. I’ve also discussed how small companies may disproportionately benefit, and that this presents an opportunity, because the maths of scaling it all may not add up for big or legacy organisations. That’s competitive advantage, folks!

However. I’ll just say that it looks really out of touch to mock concerns about AI when young people are facing a collapse in jobs and opportunity on the one hand, and being told by billionaire AI bros on the other hand that all their problems will be solved by the thing that conveniently, those same AI bros are selling.

Hey, look! It’s an article in The Economist about the problems of rising AI usage bills.

The kids aren’t stupid. The kids are alright.

4. Let’s Talk Taxes

Do I have to explain progressive taxation to The Economist?14

Let’s be charitable and assume that’s what they mean by this talking point. To which we then say... er, yes? If you create inflation (growth) but don’t control how that is distributed, the common result is wealth concentration.

The point of wealth taxes is to tax wealth, not work. It doesn’t mean that high earners couldn’t pay a bit more than they currently do. Taxes aren’t mutually exclusive as policy levers. Higher income tax in the top band is a reasonable deal, in fact, in the context of an economy that could (and should) bring back in free university and vocational education.

Personally, I think wealth taxes are a bit of a red herring—they don’t bring in enough to be transformational, but they bring in something.15 You’re going to have to also change other progressive taxes (income tax) and reform council tax in order to get enough money to have credible, transformational, spending plans.

I remember when, as a student in 2006 the UK finally paid off its World War Two debts. As a result, I was delighted to see these interviews with Mariana Mazzucato where she points out—correctly—that we can always find money to fight a war, and those kinds of long-term debts are not factored in to whether financial markets will lend money to an economy in the Global North.

Austerity was justified using an analogy (that the national budget functions like a household) that is incorrect—and the structure of long-term debt for things like war loans bears this out. Some kinds of spending are always possible, if the justification is deemed “good enough.”

The same is true of a programme to fix the UK economy. If it is credible, the money will be there.

This is why we don’t need to worry too much about the bond markets so long as we can articulate a long-term economic plan that sounds credible. Truss’s failure was being less convincing than a lettuce, and using uncosted, vibes-based tax cuts. We can do better. Much better.

If The Economist is also talking about initiatives like LVT16 (which I am not sure the Greens currently have a position on), then there is evidence that this would be a net positive for most households in the UK, representing a good way of not only making Council Tax fairer across the UK, but also working towards redistribution.

Finally, let’s talk again about wealth concentration. As another Progressive Economy Forum working paper notes, in the UK since 1980, “the market value of privately held wealth (property, physical and financial assets net of debt) has risen at more than twice the pace of the economy, a pattern repeated across the globe.” No surprise that this broadly has coincided with successive tax cuts for the rich.

Trickle-down economics, the Reagan-era idea that making the rich richer17 benefits all participants in an economy, has been pretty roundly debunked at this point. Moreover, it’s part of the reason that Liz “the lettuce” Truss’s mini budget was rejected by the markets—it’s as if somebody read Naomi Klein’s The Shock Doctrine and went, “oh yeah, great idea!”

Anyway, the paper debunking trickle-down economics that I referenced just now was the most downloaded LSE-hosted research paper ever, and is worth a read. There’s some cracking lines in the cover article, however:

In explaining why cutting taxes for the rich did nothing to boost economies, Dr Hope referred to the economist Thomas Piketty, who argues that unless capitalism is reformed, it will threaten the democratic order.

“Our results align pretty closely with some work from Thomas Piketty, that would suggest that what happens if you cut taxes on the rich is that they then bargain more aggressively for their own compensation at the direct expense of workers lower down the income distribution. So, the story of the paper then is really to do with rent-seeking among CEOs and top executives—and that increasing when you have lower taxes on the rich.”

Rent-seeking is the effort to increase one’s share of existing wealth without creating new wealth—rather like a greedy child demanding a bigger slice of the pie so that there’s less left on the plate for everyone else.

I like this quote because it also neatly explains rent-seeking. Rent-seeking crops up all over the modern economy, but not just in housing, which is a common misconception.

Those of us who were in crypto for reasons other than pure gambling18 were likely in it because the economic model of web2 encourages the development of rent-seeking platform capitalism (Varoufakis) or vector capitalism (McKenzie Wark), two different terms for monopolistic, rent-seeking behaviour. Think the business model of Amazon, Uber, et cetera—that’s classic rent-seeking.

Perhaps the most disappointing thing about web3 is that many blockchain Foundations essentially re-created the platform or vector capitalist model of coercion and centralization in their blockchain economies, something I’ve written about in conference papers presented at FCiR25 and SDLT9.

Anyway, back to the point: in an economy struggling with flat productivity, flat growth, and rising inequality, rents are problematic, non-productive capital is problematic, and it’s not just radicals that think or have thought so. Imagine my suprise that Winston Churchill of all people agrees with me:

Land Monopoly is not the only monopoly, but it is by far the greatest of monopolies—it is a perpetual monopoly, and it is the mother of all other forms of monopoly.

Unearned increments in land are not the only form of unearned or undeserved profit, but they are the principal form of unearned increment, and they are derived from processes which are not merely not beneficial, but positively detrimental to the general public.

Wealth taxes are an attempt to curtail the impact of rent-seeking. Frame it that way and they are a class of taxes addressing a very broad problem.

5. Siding With Billionaires is Pathetic

Billionaires have the money to spare.19

“But they’ll leave!” They already live extra-juristictional lives, for the most part. In so far as we have any ability to force the matter, we should say, pay your share or GTFO.

These people aren’t afraid of us, they aren’t afraid of the state, and they should be.20 A journalist friend of mine memorably recounted a discussion where a politician had said, “how do you deal with a man as powerful as Elon Musk?” To me, in the retelling, he quipped, “the SAS.”

The landlord class have always been economically distorting and non productive and this crop are no different. That’s the most charitable read I have.

The argument against wealth accumulation in the renter class is not a new one. I hope I made that case adequately in the last section.

Adam Smith, John Stuart Mill, Lloyd George and Winston Churchill have all argued that rents are non-productive for an economy, and in an environment where investment (both private and public) is acutely needed, taxing and discouraging private renting in favour of investment in other areas of the economy is hardly a crackpot idea.

Plus, if you side with billionaires over your family, friends and community, you’re a bootlicker. Sorry, but there it is.

Let’s Wrap This Up

I’m a long-time subscriber to Private Eye, and there’s a tradition in that magazine to end furious letters with, “I will not be cancelling my subscription.”

The same is true here. The Economist remains one of the few news outlets that seem to bother with fact-checking and basic journalistic standards,21 even if I generally don’t care for their worldview.

However, this is a dismal, dismissive effort that can’t go unchallenged.

In exchange for hard work, people want a nice life. They want dignity. And boy, that is a hill I will happily die on.

On May 20th, 2022 the British art pop band Everything Everything released their sixth album, Raw Data Feel. In some ways, it was a return to form. The band (unfortunately) had had more misses than hits in their two preceding records.1

There’s a track on it that hit me like a truck, Track 11, My Computer.

Some of the lyrics on the album are nonsensical even by EE’s oblique standards, and that’s because their lyricist, Jonathan Higgs, was experimenting with generatively creating them:

“With assistance from Mark Hanslip, a musician and researcher at the University of York's Contemporary Music Research Centre, Higgs developed an AI bot dubbed ‘Kevin,’ named after a recurring character in the album, to compose song lyrics generatively. Higgs fed it four different sources of information—LinkedIn's terms and conditions, the epic poem Beowulf, 400,000 posts from the message board 4chan, and the sayings of Confucius—before compiling and tweaking the results into usable material. Ultimately, the bot contributed roughly 5% of the album's lyrics and a song title (‘Software Greatman’), receiving a songwriting credit in the process” (Wikipedia)

So eccentric is Higgs’s lyrical style that I’m not actually sure whether either of the chorus’ two variations is purely his mind, or the generative model:

‘Cause you can bend, you can slide
You can push me to my limit
You can run, you can crush
Every blue boy on the planet
You’re in love with the future, I don’t know why
You’re in love with the future, I don’t know why

And you can hide, you can seek
With that ultraviolet ammo
You can sing, you can play
My ribcage like piano
You’re in love with the future, I don’t know why
You’re in love with the future, I don’t know why

The line, “you’re in love with the future, I don’t know why,” came to embody everything about the chaos that had overtaken my working life.

At the time, the Juno blockchain was navigating the tail end of the Proposal 16 insanity,2 when a governance process had attempted to seize $120 million worth of tokens, and I had spent something like 2 months 24/7 firefighting, trying to broker a compromise, talking to lawyers, dealing with a zero-day cyberattack that halted the chain, and somehow running my business.

The perils of being close to the bleeding edge and saying, “damn the consequences” had never been more real, and this song was a reminder that most of the time at the frontier, you’re surrounded by gamblers—and are probably one yourself. There’s no safety in numbers, and just because you’re where the action is3 doesn’t mean you’re collectively right.

I heard this song and realised that we had collectively believed our own hype. In uncritically believing our technology was the future, we’d committed that cardinal sin that always leads to destruction: hubris.

In 2021 we’d needed that uncritical, enthusiastic energy to launch projects, often working for free, collaborating with idealists, chancers, gamblers, founders, degens and everybody in between. The blockchain ecosystem was always more varied in its personalities, ideologies, politics and opinions than its critics gave it credit for, and everybody needed to—and largely did—pull in the same direction.

However, by late spring 2022 the cracks were showing. Projects had launched, rivalries had surfaced, money had been made (unevenly) and it was time to pony up the use-cases. At just the time when the mood needed to shift a little to be more self-critical, it had become even more hysterical.4

At that time, the Cosmos experiment with DAOs was hitting the buffers. It wasn’t Juno that poisoned the well of Cosmos—depending on how you look at it, that was either the Terra/UST collapse, or the effect of continued high interest rates,5 but in hindsight, the writing was on the wall. The summer that followed was probably the last time that most validators in that ecosystem were consistently in the green.

What this experience left me with was an apprecation of two things:

1. To get an outsized return on the energy you expend, you have to be where the action is; however

2. People where the action is are by definition gamblers, so be extra skeptical of their claims, and bear in mind their risk tolerance might not be the same as yours, or as society’s.

This is extra relevant in the current AI maximalist world we live in—to be where the action is, you’ve got to pick up your shovel,6 and head in that direction. However, I’ve seen first-hand what happens when skepticism is abandoned and wishful thinking takes over.

If that wasn’t the lesson of Proposal 16 itself, then it certainly was the lesson of everything that followed. There’s nothing wrong with being where the action is, but when things are exploding is when you need to be at your most critical, honest, and level-headed.

Perhaps that tension is obvious in the posts on this blog, I don’t know—but the chorus lyric is as relevant as ever.

Building on several posts in the last two months, I realised there was a connection between our work on decentralized governance (accounting for coordination costs in an organisation or network) and the way that AI tooling changes organisation dynamics.

So, continuing that conversation, I present an AI Law:1

The ability to benefit from non-human workers is directly proportional to your human employees’ ability to account for the costs created by the non-human workers.

The general case of this can be thought of in terms of transaction cost economics:

The value of a network can only be realised if the relationship between value in the network created by adding nodes versus the cost they incur does not mean that communication within the network becomes cost inefficient.

So let’s dive in.

Agent Sets and Cost

In my last post, I discussed the cost of adding nodes (‘agents’ in the network sense, not the ‘AI agents’ sense) to a network in the context of blockchains.

In a sense, it can be thought of as sort of an inverse Metcalfe’s Law; the idea that the value of the network scales with the number of pairwise connections.

In other words, it’s what in tech we often refer to as ‘the network effect.’

That is to say the usual formalization of the simple version is,2

Which describes the break-even point. Strictly, it’s the following, where there is linear cost growth, Cn, and A, a “non-constant proportionality factor affinity”—or, value per connection.

The issue is that this assumes value outstrips costs accrued past that point, and the break even point is only related to fixed and variable costs.3 While the original simplification was exponential, Metcalfe has vociferously qualified in later writing that it should be viewed as quadratic.

We argue there is a cost of coordination that applies to these networks, as we are in the process of formalizing for blockchain governance. This cost is accrued on a marginal basis as we add nodes.

However, this also applies to any agents in a social or organisational graph. As we saw in the last post, we can cost a pairwise connection based on the type of agent that is connected. That is, we assign agents membership of a set, and assign connections between two sets a cost.

How does this relate to the cost of AI within an organisation? Simple. Let’s rebuild the logic of our prior post but using only engineers and AI agents as our agent sets.

Let’s say our organisation is made up of nodes that are agents (A), interacting with one another. Let S be the set of engineers or staff, and L the set of LLM agents in an organisation chart. S and L are both subsets of agents.

They are disjoint sets, so a given agent or node (let’s call them n) in the org chart is in either one set or the other:

We can then see that the normal costs of communication increases (such as Brooks’s Law) are simply the description of an increase of edges (that is, connections between nodes, or lines, in a visual representation) in the graph.

The distinction we make here is simply the related cost and the value accrual.

So here is the big hypothesis or question: do AI agents generate more value than the costs they create?

This is important, as it is the intersection of Brooks’ Law and Metcalfe’s Law.4

It’s important for your organisation because at present, it is only employees (the staff set S) that can account for (pay down, or offset) these costs. Employees can generate value and account for costs (as well as create cost, of course). AI Agents (L) can only create value and create cost.5

Who Accounts for Cost?

In practice this simplified ex-ante analysis is, of course, insufficient. What happens as the system operates is that it is dynamic. Obviously cost is not uniform as in the simple model. Indeed, one Agentic LLM (L) might be best thought of as intersecting with Tasks (T), a set of background agentic tasks that each have a cost.

The intuition here is simple; that adding additional sets not only introduces granularity for the purposes of analysis, but shows how rapidly cost can scale with the addition of extra agents, given that only one set (the employees or staff, S) can offset the cost creating activities of the non-human nodes in the organisational graph.

Rather than even costing edges, for now let’s use the simple intuition that we can assign a value to the node itself (maybe this is the sum of all edge costs that are added by its inclusion in the graph, I’m hand-waving a bit to simplify things).

The logic for something like that would give us the following for calculating a break-even point, where V is network value, C is total cost, and Cn is known network costs, and we have a function f that takes a new node and returns its cost. Attempt at a formalization is in the footnote, if you’re that way inclined.6

The question then becomes, what is the cost of adding nodes? Is it linear? Quadratic? Log? Exponential? is there ever a point at which, after adding a node, C is greater than V?

Bringing this back to AI agents, is there a point at which growth in C outpaces V?

Intuitively, of course there is.

So, the question then becomes: how soon do you hit it? More nebulously, in your current distribution of S to L,7 do you hit it earlier than you would if you relied on only humans, or a different distribution of S to L?

At that point, depending on who feels the decrease in value first, and how it is expressed, we might expect to see systemic problems arise. Anecdotally, we hear about bottlenecks and burnout daily in the context of expanded AI usage, and I’m not sure this is strictly a governance problem—after all, a key part of my thesis is simply this: in many cases, governance does not scale.

My guess is that for many scaling situations, adding new nodes to a network graph follows the sigmoid function described by Metcalfe. However, in cases such as blockchain validators, and AI agents, it might be skew normal (a lopsided bell curve, left chart below). Even if the value of adding agents is sigmoid (i.e. like the governance example, right chart below), eventually cost will outpace it, depending on what that curve looks like.

To quote Jamie Hurst,

The cost of building has collapsed, but the cost of aligning organisationally has not. If anything, it's gone up. When three different teams can each produce a working solution to the same problem in the time it used to take to write a proposal, the bottleneck moves from engineering to coordination. The MR review situation is a good example: it's now easier to build a new bot than to adopt someone else's, which means cohesion gets harder to achieve, not easier. We're solving more problems, faster, and the org-level alignment work is paying the price.

Metcalfe’s Law has been used to justify the valuation of Facebook (now Meta) and other tech stocks. However, whether or not the hyperscalers’ stock price is in any way related to value is debatable. It could be (and likely is) more about animal spirits than value.

While I think the intuition that network effects are part of any scaling story is correct, it’s notable that past a point these huge companies scale as much by acquisitions as by developing their core product. In fact, with some of the main products experiencing user decline, perhaps the argument that some external costs build up and make even these systems fundamentally inefficient in terms of their transaction economics holds water.

If network scaling issues in other domains hold true—and it is a big if—then likewise we should be wary of scaling our organisations and teams.

Going Further

Except, it’s not quite as simple as outlined above. As the graphs show, there’s actually a network topology (technical graph, containing developers and agents) and governance topology (containing the employees and agents described previously).

With this in mind, we need to adapt the bold statement from the introduction to make it clear that, as in blockchains, costs generated from the techncical operations of the organisation are expected to be accounted for in the governance of the organisation. Similarly to a blockchain, in other words. Thus:

Conclusion

Hidden costs are nothing new to technical strategists and technical leaders. We’ve known about, debated and discussed these things endlessly. We usually use the jargon TCO, or Total Cost of Ownership, rather than the more economics-y terms8 I’ve used here, but the principle is the same. AI has blown traditional ways of reasoning about, and accounting for, costs out of the water, and TCO is back to the top of the list of burning issues.

The state of the art in academia seems to show a task-contingent cost curve, referred to as A Jagged Frontier (Dell’Acqua et al., 2026), which means that the benefits of AI usage are felt unevenly.9 On average, AI use, whether supervised or not, was found to always improve output when “within the frontier,” and result in worse results outside of it.10

Taking the ideas from our last post, we could build this cost-curve based on sets of tasks, where each have different costs. This would show uneven cost of AI use, and how it contributes to total organisational cost. Figure out value per task (easy, right?)11 and you could overlay the two.

The question not addressed in the study is a longer time horizon—whether deskilling, or other factors would change this result over time, shifting the cost and value curves. Within the bounds of the study however, the result appears clear—some amount of AI use, on average, will result in better outcomes on within-frontier tasks. Any negative effects might simply be a cost of doing business in the AI age, where a black swan cost or risk generated by an out-of-frontier task could sink your project or business at any time.12

Variations on such risks have always existed.13

The good news, then, is that understanding and managing TCO was already most of the ballgame, just as things like TDD and good engineering standards were. So nothing changes, right? Because we all had everything figured out and running smoothly before the AI wrecking ball came along…

NB: An earlier version of this post did not reference Jamie Hurst’s blog post. The post was updated to include a quote.

Apologies for anybody that was after a ‘light’ piece this week—instead it’s a work-in-progress piece of academic work. For a while I’ve been trying to formalize some way of structuring the agent types in a blockchain network to show how risk, cost and security are felt, and for now, I’ve settled on pursuing an angle that involves weighted sets.

In this article I’ll talk about two ways of looking at agent sets for analytical purposes: weighting the nodes in the set, and weighting the edges between nodes.

First however, we need to talk about how these networks are structured. Based on qualitative research, industry experience and an extensive review of the literature, we advanced a model that operates on two axes—the technical and the social.1

1. Network Topology and Governance Topology

In our work, we model a network as split across two axes:

• Network topology, the physical and technical structure of the network.

• Governance topology, the structure of decision making power in the network.

We summarise this in our papers as a split between the ‘protocol’ and the ‘governance of the protocol.’ It’s a useful way of looking at socio-technical systems, especially when it comes to how externalities are felt.

We hypothesise that many of the external costs generated by the network topology are assumed to be accounted for in the governance topology—however, in practice, they usually are not.

Finally, these axes—or perhaps ‘layers’ would be a better term—are stratified, in the sense that the governance of the network always takes precedence.

2. A Weighted Sets Model of Decentralization

Let’s attempt a formalisation that shows the interaction of agents as they account for the cost of security2 in a network. Following the reasoning of Budish (2025) we assume:

1. Adding a single node to the network always has a linear cost as the best-case

   scenario; thus, the weighting factor is always at least 1.

2. The combined cumulative value of all nodes multiplied by their weights adds up to the combined fixed cost of security for the network.

This would not be the total cost, due to the presence of externalities3—we hypothesise that there is a negative externality associated with the cost of co-ordination in both the case of a network and a governance topology that is not accounted for on most existing ledgers.4

Excluding exogenous externalities as beyond the scope of this piece, from the perspective of a single agent submitting data to the consensus and finality process, we can model the structure of the cost of finality in terms of its locus in the subsystems and agent personas in a system.

Let’s assume, for this simplified model, that agents can be in only one subset. Thus, we might imagine the network topology of a network as a set of subsystems or nodes, S. However, different nodes or subsystems in this entity graph, for example, validators V, software implementations I, a Foundation, F, or data centres D, have different weightings for their centrality. Another way of thinking of this is in terms of their centrality to the integrity guarantees of the system in operation.

We can imagine a formalization of weighted centrality that is as follows:

1. A weighting function:

2. A function (sigma) that maps subsystems (S) to their subset or partition:

Where P is a subset, or partition of S.

This means we can use a weighting function (w) for a single subsystem (n):

To confirm the total weight sums as expected, it is possible to sum the weighted sets:

This same exercise can be completed for the governance topology of a network. As stated before, with ‘real’ values, we would expect this sum to add up to the (fixed) cost of securing a network at a given point in time.

The questions then are:

1. What are we weighting? There are two options:

   1. The nodes themselves

   2. Their relationships (which would mean we actually need to uniquely weight the edges between distinct sets, e.g. Validators (V) and Data Centres (D))

2. How do we represent externalities?

Asking the first question raises some questions about what exactly weighting the nodes tells us.

3. Visualising Weighted Sets as Cost Curves

The goal here is to look at all the potential places in the network that affect the integrity of our assets,5 and apply a weighting to them, which we can later express as a cost.

This shows our risk exposure as a transacting agent within a network to other agent types, as well as the relationship to the cost of security. In the next section, we will describe how this might be calculated such that it would represent a cost curve.

To model a public, permissionless network, we simplify the topology down to only three components: client implementations (1), validator set (100), and then all other subsystems, which we assume add up to a further 200 nodes—and to whom we apply an equal weighting.

The single client implementation is the first point of centrality, followed by the hundred validators, followed by the remaining subsystems (in this case, we allow for only one type—we could easily say it simply represents stakers).

Currently these charts use only arbitrary numbers, but if we exempt client implementation in the case of an example system such as a Cosmos blockchain or Sark network,6 we could model the effect of a ledger as single point of failure, versus a system where a higher level of trust in a single agent (a Porter) was required.7

‘Porters’ are simply the in-system name for relays, to which an asset is ‘tethered’ for its integrity. The transactable asset in Sark is the USO, an Unforgeable, Stateful, Oblivious asset8 and can only be updated by a single relay, which is identified in the asset’s data payload by the identity of the Porter which is allowed to update it.

4. Weighted Nodes vs Weighted Edges

This weighting is all very well, but it’s still not telling us much. My first intuition was the weighting could result in a gravity-weighted graph that would show the relative centrality of a Foundation or controlling entity.9

However, I think the place to take this next is to maintain the partitioned sets model, but abandon weighting the node itself, and add an additional group of sets: the edges between nodes (i.e. agent sets). By doing this, we can later use the edges to weight the node based on their relationships.

Then you can start to model value of adding an agent, and cost of adding an agent. Think of it in terms similar to Metcalfe’s Law—that adding an agent has a value, related to its set, as well as a cost.10 Of course, this is unlikely to be linear. In a BFT system, the 4th node is the most valuable systemically, and we would assume a long tail of diminishing value accrual thereafter.

For simplicity, you would weight the edges between nodes in the same sets equally (even though concerns such as voting power, jurisdiction etc might change the weighting). In practice, a Foundation might delegate a hundred USD to one validator, and a million to another, but here, we would assume equal weighting of edges (pairwise, symmetrical connections).

In an example, the subset of edges, E, should be the same whether a V (Validator) or F (Foundation) is the source. Thus, given a function that can return a weighting for an edge, given its source and target, f, the value of a node is the sum of this set of edges.

This would allow you to simply sum the value of all the edges to a given node, giving the node a concrete value.

In this simplification, it’s bi-directional, but that naive weighting might not work. At the end of this section, I’ll add some thoughts about how this relates to cost, but the intuition should be that it operates similarly to Metcalfe’s Law, and the visualisations in the previous section are simply the cost curves.

Let’s think of a visual example. Users can hold one unit of an asset—a USO in the case of Sark, and a whole token of some nominal cryptocurrency.

How would we put a number to it? For a dummy example, the intuition would be to weight the edges between two sets as 1/(number of nodes that have responsibility for integrity) in the case of integrity.

Consider the case of a Porter from the point of view of a USO asset holder—to update that asset, they (dark blue) need to talk to only the Porter (light blue) specified in the asset by either the issuer or the last holder. Therefore, the weighting of their edge with that Porter is 1 (1/1).

Now consider the case where we have a cryptocurrency typical of a public, permissionless network. It can be updated by sending a request to any Validator (in fact, any RPC, but let’s ignore RPCs for now in this super-simple example). We omit RPCs because we care about the subsystem responsible for integrity—the valset. RPCs are only for reads. Assuming, as in the prior examples, a valset of 100, the user’s dependence on that validator is 0.01 (1/100). The edge could be weighted by that value.

Here’s a simplified version, with 20 Validators (light blue) and a single user (blue), with a weighting of 0.05 per edge. The Validators are all assumed to be 100% reliant on the Foundation (green) for their existence/integrity (profitability). In reality, they’re somewhat dependent on users, but we base this on our fieldwork, where Validators rely on Foundations and generally don’t find gas, MEV or other incentives to be key in the long-run. However, this isn’t quite right. They’re dependent on one another too. Here they’re in a full P2P mesh.

To bring in cost, and weight the set in the context of the whole system, we need to change that value of 1 (even though in the above examples, it wouldn’t tell us much, since only validators and users are represented). Then, assuming the weighting value for each set of edges between agent sets was scaled appropriately to the total fixed cost in the network, the edges would be weighted according to cost.

In the case of adding an agent, there is a marginal cost to adding the agent, but from the perspective of a marginal transaction, this is likely to be felt as a fixed cost. The intuition here is simple: if the edges’ weighting is, in a sense, the fixed cost, these visual representations form a basic hypothesis as to whether these systems can scale, or not.

Even the existence of an edge is significant for describing the cost of system-level coordination. What it shows is that the system’s topology in terms of connections matters—in both the network topology and governance topology.

It’s easy to see that naïve decentralization (or performative decentralization, also known as ‘decentralization theatre’) along the lines of “more nodes is better” simply leads to an increase in cost. The question is—the fixed cost can be seen clearly, but if there is an external cost, then it could be exponential, depending on the node graph. How do we represent it?

Conclusion

As I suggested in the section before, it seems likely that modelling edges might be more informative. The key reason is this: adding another edge always increases the total cost, from the perspective of the user, even if the weighting is small. As you’ve probably guessed, if this logic holds, it will be coming in a paper soon.

Acknowledgements: although this work is at a very early draft stage, I think it’s valuable to try and generate some discussion, so I am putting it out there. It builds upon many discussions at UCL and specific sessions on modelling externalities and scaling permissionless networks with Prof Tomaso Aste and Dr. Geoff Goodell.

I should note that Tomaso and Geoff would likely characterise the current state of my work as not far above the ‘shower thought’ level, so no endorsement is implied or indended—merely that I value the conversations I’ve had.

Anyway, at the time I tried to distil what I learned about agile into a post that was a 1-minute read. It wasn’t bad.

Since, I’ve worked for a startup that was later acquired, contracted as a fractional CTO, done numerous digital transformation and due diligence projects, been the technical lead1 for a blockchain, started a company,2 and launched two blockchain protocols.

So what have I learned since?

Funnily enough, the original post mostly stands up. Probably the main thing I’d change is making the importance of trust3 even more central to both facets of delivery. I’d also emphasise feedback loops. To some degree, what you’re really trying to do with agile is:

1. Identify the desired outcome

2. Develop systems around feedback loops to iterate to that outcome

In a sense, this terser definition is even better than my past one, as it doesn’t rely on specific tools, approaches or conditions; it is generic advice (in the good way!).

As sharper readers will have spotted, as this is a generic approach, it won’t have any real impact unless the right outcome can be identified. That’s a very deep rabbit-hole, and outside of the scope of this post.4

However, in the spirit of the original post, I’ll have another crack at it:

Tech side

What?

• Reduce cost of change

• Reduce cost of failure

• Fail fast

• Identify and solve the right problems

How?

• Trust

• Staff agency

• Experimentation

• Automation

• Iteration and building feedback loops

Why?

• Deliver faster

• Be able to scale solutions

• Make money

• Save money (either absolute terms or TCO)

• Address correct outcomes

People side

What?

• Meet user needs

• Reduce staff churn

• Improve ability of staff to address relevant business problems

How?

• Diversity of thought and experience5

• User research

• Building trust

• Building agency

• Iteration and building feedback loops

Why?

• Deliver smarter

• Solve the problems people care about

• Make money

• Save money

• Address correct outcomes

In both the ‘why’ sections, I’ve made a clearer link to business outcomes; in both the ‘how’ sections I’ve foregrounded trust and agency.6 I’ve also dealt with staff retention, hiring, churn and output in various guises in my consulting work, and so that has to be added to the ‘what.’

In both cases, I’ve seen a lack of agency expresses itself in both an inability to identify the most important problems, and then a lack of agency to actually tackle them.7

Finally, I’ve brought in the points about outcomes and systematizing feedback loops. Even within a business, you need to think about scaling an approach, and that’s what systematization does—allows you to take a playbook and apply it to different systems, processes or problems within a business to drive an outcome.

Indeed, it’s why, during my time in a transformation team, I worked on projects that were digital-first, in-store, in funeral care (both physical and digital) in data, and in hardware. The same set of analytical tools and system thinking can lead to problem solving in heterogenous domains. It’s basically what I’ve been doing since, both in my day job and my academic work.8

So with this week’s theme of “Can you spot a disordered personality?” in mind, let’s revisit some definitions and then talk about them the context of workplaces.

Obviously:

1. As in previous posts, I’m talking about the non-criminal/non-clinical end of the disordered personality spectrum here.

2. I’m not encouraging vigilanteeism—I’m mainly writing this so you can spot the signs and minimise these peoples’ impact on you.

The Field Guide

I’m adapting some content here from my review of Nate Silver’s book On the Edge, since I quoted a several academic definitions of non clinical/criminal Dark personalities (Narcissists, Machiavellians, and Psychopaths).

That post also covers the crossover (or not) with the Dark Triad, as originally explored by Paulhus and Williams (2002). We’ll reference that paper a lot in the next section(s).

A later (2024) article by Walker and McCann has some nice, terse definitions of the disordered personality types:

Narcissism

As a subclinical personality domain, narcissism covers a spectrum from mild to extreme (Miller & Campbell, 2008) and comprises two facets: grandiose and vulnerable narcissism. Grandiose narcissism is characterized by grandiosity, self-confidence, and exploitation of others with a tendency to rely on self-internal validation (Dickinson & Pincus, 2003; Zhang et al., 2017). When threatened, people with high grandiose narcissism may blame and devalue others while refusing to acknowledge their own weaknesses. Vulnerable narcissism is characterized by grandiose fantasies, oscillations between self-love and self-loathing, a fragile sense of self, a reliance on external validation for self-esteem maintenance, and hypersensitivity to negative feedback (Wink, 1991). When threatened, people with high vulnerable narcissism become defensive and resentful, and may show aggressive outbursts toward others (Wink, 1991).

Walker et al. (2022) in a different paper on faking good and bad on tests have several useful definitions, which I quoted in my previous post. For Narcissism, I just want to re-quote one point on what happens in the case of criticism or conflict (my emphasis):

Individuals high in vulnerable narcissism are characteristically hypersensitive to negative feedback and tend to act aggressively when their sense of self is threatened (Wink, 1991). Though grandiose and vulnerable narcissism share several core characteristics such as an antagonistic interpersonal style, self-importance, entitlement, and hypersensitivity to criticism (Dickinson & Pincus, 2003; Krizan & Herlache, 2018; Weiss et al., 2019), there is clear evidence that these two facets are distinct (Krizan & Herlache, 2018; Miller, Vize, Crowe, & Lynam, 2019; Walker et al., 2021).

The entitlement point is something that I think comes across a lot when you listen to these individuals talk—particularly in the political sphere.

Machiavellianism

Subclinical Machiavellianism is derived from the philosophical writings of Niccolo Machiavelli, a political advisor to the Medici family in 16th century Firenze (Christie & Geis, 1970). Machiavellianism is characterized by goal-focused manipulative and callous social interactions, including the use of long-term strategic planning to delay gratification for better rewards in the future, questionable morals, and a cold, cynical world-view (Christie & Geis, 1970; Furnham et al., 2013).

Psychopathy

Subclinical psychopathy is characterized by superficial charm, pathological lying, and lack of empathy, conscience, and remorse (Cleckley, 1951; Hare, 2003), existing on a continuum in the wider population (Berg et al., 2013). Psychopathy measures, such as the Levenson’s Self-report Psychopathy Scale (Levenson et al., 1995), were developed to measure the two-factor structure of psychopathy proposed by Karpman (1941). This two-factor structure comprises two related but distinct factors differing in their etiology and symptomology: primary and secondary psychopathy (Hare, 2003).

Both facets are typified by indifference to one’s own and others’ emotions, underpinned by an antagonistic interpersonal style (Miller & Lynam, 2012)…Egocentricity relates to interpersonally manipulative and antagonistic characteristics associated with perceived low social responsibility (Christian & Sellbom, 2016; Sellbom, 2011), callousness relates to lacking empathy and remorse and is associated with cold-heartedness, lack of remorse, and low empathy (Anderson et al., 2013; Sellbom, 2011), whereas antisocial is related to impulsivity, and antisocial behavior (Brinkley et al., 2008; Sellbom, 2011).

As in my previous post, I think it’s worth discussing the difference between Primary and Secondary psychopathy, so I’ll again quote Walker (2022):

Primary psychopathy is characterized by: (a) a lack of guilt and remorse, with elevated levels of callousness, manipulation, and socially desirable behavior (Hare, 2003), (b) having superficial affect (Casey, Rogers, Burns, & Yiend, 2013), and (c) deficits in affective empathy (Wai & Tiliopoulos, 2012). Primary psychopathy is also associated with lower levels of fear (Patrick, Bradley, & Lang, 1993) and lower indications of repentance (Hare, 2003; Lee & Salekin, 2010).

Secondary psychopathy is characterized by higher levels of antisocial behaviors such as aggressiveness, impulsivity, and anxiety (Levenson, Kiehl, & Fitzpatrick, 1995). These types of characteristics are potentially a result of experiencing strong emotions which are unable to be effectively regulated (Hare, 2003). Additionally, individuals high on secondary psychopathy have been shown to possess guilt and fear responses not typically observed in individuals high in primary psychopathy (Lykken, 1995; Wallace, Malterer, & Newman, 2009).

A Narcissist and a Psychopath Walk into a Bar Their Place of Work

To discuss some of the heuristics you can use to validate a hunch, the study contained in Paulhus and Williams (2002) has some useful data and discussion, so I’ll quote from there. It contains a lot of references to prior studies on the different personality types, which are interesting if you really want to go down the rabbit hole!

First up, I’ll note that all three Dark Triad personalities tend to lie, a lot. So if you have somebody that often says things where you think “that can’t be true,” or “that must be an exaggeration,” then perhaps your gut is telling you something.1

The other things they have in common are a lack of empathy, callousness, and a core of aggression (either passive or straightforwardly aggressive).

I’ve noticed weird “fight or flight” impulses tied to meeting people like this, including a time when I was introduced to an older gentleman at a wedding. I couldn’t tell you why—but I felt he was squaring up to me, for no discernible reason. I’m certain he was a disordered personality.2 Why he gave me that feeling, I’ve never been sure.

Before we go on—a pop quiz: What are Mrs Bennet from Pride and Prejudice, Homer Simpson, and Wallace from Wallace and Gromit? (Answers in the footnote)3

Narcissists

Finding Narcissists is usually pretty simple, especially grandiose ones. They want to be the centre of attention and at the centre of gravity for any social network.

By definition I’m talking about non-clinical cases (both for Psychopaths and Narcissists). These are, after all, people that can hold down a job and go unnoticed as problem personalities by many people.

In my experience, drama seems to follow these people, but they often also manage to stand above it. At the very least, I’ve noticed with Narcissists, they don’t so much stand above it, as people will go out of their way to excuse their behaviour, usually (I think) out of fear.

A more charitable read could be that somebody’s behaviour is only going to improve if you cut them some slack. Maybe they’re going through some stuff. Disordered personalities drive a truck through this loophole in human social norms.

If there’s somebody in your workplace that is at the centre of everything, that everybody speaks well of, but also everybody seems to tiptoe around—bingo! You’ve found your candidate.4

A heuristic that is useful is the degree of delusion for the three personality types. Paulhus and Williams (2002) describe that Narcissists are the most self-involved, and least grounded in reality, “narcissists have a strong self-deceptive (i.e., low insight) component to their personality (Raskin et al., 1991; Paulhus, 1998),” followed by Psychopaths, especially “clinical range” ones—Machiavellians, by contrast, are more “reality-based in their sense of self.”

What this meant in studies that surveyed student populations was that a good (but not perfect) predictor of Narcissism was where respondents over-estimated their own intelligence, “narcissists and, to a lesser extent, psychopaths tended to overestimate their intelligence… whereas Machiavellians did not.” (Paulhus and Williams, 2002)5

Back to Mrs Bennet—her own desire to be at the centre of her children’s matchmaking results not only in delusional behaviour but also from a goal-oriented perspective, near catastrophe in actually reaching her desired outcome.6 Entitlement, lack of empathy and social awareness are her characteristic traits.

Another heuristic I’ve seen is the following, that only Narcissists can say: “You hurt me! It’s your fault!” as well as “I hurt me! It’s your fault!” Especially with the victim complex inherent in vulnerable Narcissism, there’s little room for them taking responsibility for their actions, and lashing out or looking around for someone else to blame is the norm (look at Trump over Iran).

Grandiosity and drama resulting in sub-dominant outcomes for Narcissists is something I’ve seen over and over again, and was recently played out in perfect fashion with Fiona (a Narcissist) on the BBC series The Traitors, or when Andrew Tate revealed his location while arguing with Greta Thunberg, leading to his arrest.

Final point—be a little bit wary of anybody that’s a self-diagnosed autist (I say this carefully, with diagnosed autists as friends and family). They might be a Narcissist using self-diagnosis to shift the blame onto a different condition with some superficially similar behaviours. The strategy makes sense—even diagnosed autism potentially has a limited comorbidity with vulnerable narcissism. Thus, it’s important to stress that they are different.

In my experience, my autistic friends occasionally display a superficially similar but distinct pattern of behaviour—things like slight selfishness in certain situations where they miss social cues, or a tendency to do most of the talking when a subject comes up they’re interested in (pot kettle black, I know). I can think of examples like a friend whose legs I had to lift up to clean around (more than once) as he lay on the sofa, but jumped up to help the second he was directly asked. Think of Wallace from Wallace and Gromit.

Simply put, in the words of this post, autists aren’t a—holes:

“The distinction here is between the person who’s often rude and the one who’s often, or at least occasionally, nice…Narcissists crave admiration, whereas autism desperately maneuvers for acceptance…The difference between autism and narcissism, [is] that one implies a high degree of non-sensitivity, whereas the other implies one of insensitivity; the former doesn’t know and the latter doesn’t care. The narcissist displays patterns of indifference and cruelty, whereas the autistic individual may feel remorse for some social blunder.”

For your friends and family, it’s easy to identify if they mean well. The difficulty of distinguishing patterns of behaviour for people you don’t know well, or sorting the claims of the self-diagnosed, is perhaps at the heart of the reason I think Nate Silver spent so much time musing on autism in his book—when he should (in my opinion) have been investigating disordered personalities.

When Musk did a ‘Heil Hitler’ at Trump’s inauguration, some clowns were attempting to excuse him based on his ‘autism.’ Being autistic doesn’t mean you can’t be a Narcissist, Psychopath or Dark Triad, just as being autistic doesn’t mean you are a Narcissist, Psychopath or Dark Triad. I don’t understand why this is hard to get.

Psychopaths

For psychopath-hunting, as documented by Babiak and Hare7 and others, if you’re in a large corporate, it’s a good bet to look at your boss and start going up the org chart. You’ll find one soon.

I remember being in an all-hands to be introduced to our new exec member and watching in disbelief as he stood at the front and said, while presenting a slide deck about himself (I paraphrase here due to brevity and length of time since then), “my PA said I should put a picture of my family here, before the dogs, but…” he then advanced to a slide of his dogs.8

A good heuristic for finding Psychopaths is that childhood delinquency predicts higher non-verbal IQ and SRP scores. Again, it’s not perfect, but it turns out, antisocial behaviour is a pattern, even if confined to relatively harmless things.

Psychopaths also lack anxiety. Think about the stories of that boss that had a major relationship issue, then appeared at work fine the next day, or gave out cake just before announcing redundancies.9

Although it doesn’t help you spot one, as mentioned above, Psychopaths are typically better at non-verbal communication cues, which is at least a way to distinguish them from Narcissists.10 Both are “associated with extraversion and openness.” (Paulhus and Williams, 2002)

If you need a personification of the (non-criminal) Psychopathic personality, then read almost any Richard Morgan novel—Takeshi Kovacs (Altered Carbon novels and TV series) and Hakan Veil (Thin Air) are psychopaths with a side of bipolar disorder, and Market Forces and Thirteen are just portraits of a psychopath.11

If you can’t spot a psychopath, unfortunately some criminal psychopaths can spot their victims just by walking gait. A 2013 study showed psychopaths were adept at spotting people vulnerable to victimization in this way.12

In the Paulhus and Williams paper, they find men display more of the Dark Triad traits, and that the strongest inter-correlation is between Psychopathy and Narcissism—which makes sense, since grandiosity is also to some extent a Psychopathic trait.

On which final point, it is possible to distinguish the three Dark Triad disordered personalities by their degree of self-aggrandisement:

On two objective measures, narcissists exhibited the most self-enhancement, followed by the psychopaths…In contrast, machiavellians showed no sign of self-enhancement. This difference is consistent with previous evidence that Machiavellians are more grounded, or reality-based, in their sense of self (Christie & Geis, 1970), whereas narcissists have a strong self-deceptive (i.e., low insight) component to their personality (Raskin et al., 1991; Paulhus, 1998).

Although I don’t have to line manage any more (and haven’t for some time), I’m still in the habit of trying to distill key points here and there from books that I’ve read. Partly (as I’m sure others will sympathise with) this is so that I don’t forget them, or can refer to them later, and partly it’s because I still do a little bit of mentoring and career advice and these notes can be useful in those cases.

One of the most interesting chapters in Nate Silver’s 2024 book On The Edge is the middle chapter on risk-taking behaviours. Something that I’ve observed a lot when looking at the careers of others is that they don’t take enough risks. I’ve never particularly thought of myself as a risk-taker, but with the benefit of fifteen years of hindsight, it appears that, compared to the average, my perception might not be correct.

While reading the book, in fact, I realised that on some level, much of my advice to folks (particularly those at the start of their careers) could be summarised as, “take on more risk,” with a side of “but make sure that it’s calculated.”

To make sure that you’re making a good calculation here, a good rule of thumb is to figure out what the upside of a given choice is—the terms I tend to think of here are simple: money or learning—and optimise for what matters to you most at the time.

I’ve taken jobs that were more about the money, some that were more about the learning,1 and some, like my second engineering job at the data consultancy Swirrl,2 which rather fortunately offered both.

As Erik Deitrich points out, when you take a job, you tend to get more benefit than the company that hired you, especially if you’ve chosen a new workplace where you will upskill, but that over time, this learning levels out, and the relationship becomes that your employer benefits more than you do. This is why it’s important to have clear success and exit criteria in mind to govern when you will move on to your next challenge, a mentality he describes as “Always Be Leaving.”

I’ve always encouraged folks to have a rough five year plan3 and to have some kind of heuristic for making decisions. In Scott Hanselman’s Scaling Yourself talk, he references a quote from Christopher Hawkins, which I paraphrase: “if it isn’t making me money, I drop it.” He then flips that to put his own primary goal in its place, “if it isn’t helping me spend more time with my family, I drop it.”

This kind of heuristic is continually useful, for evaluating both decisions in the large (do I take this job) and the small (do I take a meeting in person rather than online).

But anyway, back to risk. Silver’s 13 habits aren’t all strictly relevant to thinking about risk in your career, but some of them are (I’d say 1-3, 5, 8-10, 12, and 7 is crucial—these are all in bold):

1. Successful risk-takers are cool under pressure.

2. Successful risk-takers have courage.

3. Successful risk-takers have strategic empathy.

4. Successful risk-takers are process oriented, not results oriented.4

5. Successful risk-takers take shots.5

6. Successful risk-takers take a raise-or-fold attitude toward life.6

7. Successful risk-takers are prepared.7

8. Successful risk-takers have selectively high attention to detail.

9. Successful risk-takers are adaptable.

10. Successful risk-takers are good estimators.

11. Successful risk-takers try to stand out, not fit in.

12. Successful risk-takers are conscientiously contrarian.8

13. Successful risk-takers are not driven by money.

If you want to read his full reasoning, pick up the book—obviously this is just a tiny summary for the purposes of this article. I’d also note—although this is a subject covered elsewhere in my review of the book—that many of these behaviours are also present in non-clinical psychopaths. Oh well.

I realised that I’ve never written about one of my favourite books. Skunk Works is the story of the Lockheed Martin Skunk Works, the top-secret engineering shop that produced (in secret, and sometimes under budget) some of the Cold War’s most incredible pieces of aerospace innovation—the U2, the SR-71 Blackbird, and the F-117 Nighthawk.1

If you’re interested in an insight into that period of defence history it’s a pretty fascinating read, but it’s equally instructive as a book about management and organisational design.

The team could be characterised as having high agency, but high responsibility, and the amount of initiative that designers, engineers and assembly staff used is continually surprising to a reader used to the micromanagement inherent in the modern corporate workplace.

Managers were consistently able to provide what I’ve seen described as ‘air cover’—insulation from interference by stakeholders and bureaucracy—while delivering against the real objectives of a project and the outcomes that mattered.

The book covers what (necessarily for the narrative to work) is framed as a golden age, from shortly before the development of the U2, to the operational deployment of the F-117 during Operation Desert Storm.

Anyway, you should absolutely read the book, it’s great, but there’s a passage in it which I go back to a lot, have described to other people, and I gather is valuable.2 So here goes.

In it, the author Ben Rich speaks to his boss, the legendary aeronautical engineer Kelly Johnson, and says he’d like to attend a course at Harvard Business School. Kelly writes him a “glowing recommendation” and gets Lockheed to pick up the bill, even as he declares it “a waste of time,” giving Rich instead three concise rules for leadership:3

1. “You don’t need Harvard to teach you it’s more important to listen than talk.”

2. “You’ll never make the grade unless you are decisive: even a timely wrong decision is better than no decision.”

3. “Don’t half-heartedly wound problems - kill them dead.”

“That’s all there is to it,” Kelly concludes. Of course, Rich goes to business school anyway, finds it somewhat useful, but ultimately values Kelly’s argument enough to include it in the book.4

Hopefully you find these rules useful.

Over the last few years I’ve worked on, and managed the development of, a lot of open source code. As you might expect, it’s primarily been hosted on cloud platforms (mostly GitHub, occasionally GitLab).

I had a realisation last month—that my relationship with open-source code was changing. My internal cost/benefit calculus was shifting. At a minimum, if you’re iterating towards an MVP, it should no longer be in a public repository.1

For most organisations, this would be a given anyway—just because a competitor might not understand what they’re looking at, why take the risk? Still, the chance of that code being made comprehensible by an AI assistant is pretty strong, so this scenario has gone from a poor strategy to suicidal strategy.

Okay, in all likelihood, you’d have built in a private repo anyway, but is that safe?

Although GitHub say that they don’t train their AI models on organisation private repositories, they’ve made no such claims on individual accounts. Indeed, any attempts to gain clarity on the matter using their issue tracker et cetera have been met with crickets. It’s been the subject of threads on Reddit and Hacker News, with some users openly not trusting the public statements given by the company. This comprehensive post by Simon Williamson digs into the matter in depth and still can’t reach a definitive answer.2

This does ultimately mean you’re putting your faith in a large organisation (Microsoft) with hugely asymmetric power over you to both a) not act maliciously and b) not act incompetently. Quite a big ask.

After all, from Microsoft’s point of view, they’re not in the business of stealing the IP of nascent ideas—they’re happy to buy them—so you’ve got to ask the question, how does it even get back to them if a) or b) above is violated, and your IP is leaked via Copilot?

This may be paranoia, but it’s worth considering. In my previous post on Hard Problems, I pointed out that in the age of AI coding agents you need hard problems with barriers to entry. A good way of ensuring the barrier to entry you thought you had (your unique idea, GTM and IP) remains unique is to apply some adversarial thinking to your development stack.

This is all rather bizarre for somebody that’s worked in crypto for the last 5 years—where your DMs blow up if you release anything that isn’t fully open-source, and your competitors brazenly just fork your code and deploy it—but I can see myself in the future taking not only steps back from open-sourcing system components that are part of products we are building, but even going so far as to host them on company Git servers instead of cloud platforms.

In my life as a consultant I talk a lot about TCO—it’s the reason most businesses should be using these cloud tools, rather than rolling their own—but by the same logic, I now find many of my own projects falling on the other side of that fence. There’s a cost to risk, and I think it would be naïve to not factor that in.

One changed line in a cloud platform’s Terms and Conditions that gets missed, a malicious or incompetent move, and you’re toast. It’s a Dark Forest now.

Watching Inside the Manosphere the other night,1 I was struck by the cognitive dissonance of some of the personalities interviewed.2 They still saw themselves as victims, plucky rebels fighting The Culture, even as their avatar sat in the Oval Office. Somehow, despite having won by any reasonable metric, they still saw themselves as the victims, because everybody else still does not agree with them.

This single-minded fixation on being right and being the centre of the universe is a classic Narcisstic trait.

It’s also how the politics of self-interest can rapidly become the politics of oppression, authoritarianism, and silencing dissent.

However, and here is where I might make myself a little unpopular in my own political camp—there’s also a grain of truth to their original complaint about being the victims of The Culture. Ironically enough, the reason is simply this: the Left (however we define it) has its fair share of narcissists too. Of course it does.

Like narcissists on the right, they are too busy thinking of themselves to start looking for win/win solutions.3

This has always been the danger of identity politics. Unlike the politics of community and collective vision and effort, it is atomised, and on some fundamental level, essentially narcissistic. At its worst, it enforces a zero-sum view of the world that sees everything in win/lose terms rather than allowing for the possibility of win/win.

No surprise then, that the far right has decided to fight fire with fire,4 and as events have shown, are much better at playing the game than the Left.

As others have noted, a plurality of opinions are required in order to sustain a win/win equilibrium. However, there’s also a flip side to this in order to avoid the politics of narcissism. Taking responsibility for their actions is something narcissists and psychopaths are both bad at, but in order to have your opinion, it also needs to be hedged against the consequences of using it.

Being ‘cancelled’ is one of those—a social enforcement of a behaviour norm. Good manners cost nothing, as the saying goes, and the people that think they can have their opinion without also respecting others shouldn’t be surprised if what goes around, comes around.5

Proof of Ownership, or PoO is a new consensus paradigm for a stablecoin-first future. It comprises 3 main components, with the goal being to ensure an endogenous token (a so-called ‘gas token’, to many) retains value even as a stablecoin is the main transactable asset on a network.

1. Only wallets with a valid soulbound or wallet-bound (and potentially even time-bound) NFT can transact. If your stablecoin is private-by-design (it should be) then the creation of this should be a zero-knowledge event where only a proof is written to the ledger. The existence of this proof is locally required at the time of issuing a transaction by a valid client, but it not audited centrally.1

2. All transaction fees (as well as gas) are paid to the network in the endogenous token. Transaction fees should likely be stable (but one would envisage greatly cheaper than the equivalent fees when e.g. using the Visa or Mastercard rails).

3. Optionally, some amount of endogenous token is staked or bonded to a validator before a wallet can transact.2 An antehandler3 filters all transactions that would increase a validator’s share of VP above a fixed sensible point (e.g. 5%, if you assume a valset of 20). Another option that would reduce the governance overhead would be to have all vals be within +/- a couple of percent of the average VP in order to be eligible for stake bonding.

This system requires an oracle, since the value of the NFT should be stable relative to the exogenous price of a real-world currency (let’s say the dollar, if you envisage a dollar stablecoin, but you should probably denominate it in something else unless you are (a) American, (b) an American company).

Thus, at least in principle, the value of the NFT is stable.4 You might additionally consider making them bound to wallets and non-transferable. Assuming an HD type wallet, you would need one for each sub wallet.

Although this is not the goal of the system, you could imagine AML/CFT or KYC measures occurring at the point of NFT sale.

In fact, it is possible to use this general scheme on top of say, a Tendermint style blockchain and make it work with a Sark payments layer.5

Relays (Porters) would have to take additional metadata and verify NFT ownership on-chain before saving transactions to their local transaction trie, but apart from that, the principle should work.6

If assets needed to be transacted more than once (for example if they represented an RWA and not a payments primitive, like a stock holding), then you could in principle continue to use a system like Sark with multi-hop updates and saturate the full provenance of an asset to determine its history in the case of regulatory need. Let’s call this extended protocol PoOP, or Proof of Ownership and Provenance.

Having typed this whole thing, and despite the date, I’m still totally unsure if I’m serious or joking.7

💩

I came across Evans F. Carlson a few years ago and was fascinated by his story. During WWII, the US military established what was essentially the first US special forces unit,1 the Marine Raiders. They created two Battalions, and gave their commanders extensive leeway in their organisation and tactics. Merritt Edson was given command of the 1st, organising it along the lines of existing unit doctrine, and Carlson the 2nd.

Carlson structured his unit more unconventionally than Edson. As an observer, he had seen the Chinese Communists fighting during the Chinese Civil War, and had been surprised by their effectiveness. He took from this two lessons—the importance of an éspirit de corps based on co-operation, and their size of unit organisation, which was typically 3 fighters:

Carlson used egalitarian and team-building methods: he treated officers and enlisted men with minimum regard to rank as leaders and fighters, gave his men “ethical indoctrination,” describing for each man what he was fighting for and why, and used the Chinese phrase “Gung-ho!” as a motivational slogan which he learned from the Communist forces during his years in China. He also eschewed standard Marine Corps organization, forming six rifle companies of two platoons each, and innovating 3-man “fire teams” as its basic unit. (Wikipedia)

The best thing about this story is the etymology of “gung ho,” which became a war cry for the US Marines as a whole after a propaganda film of the same name about the Makin Raid.

Carlson explained in a 1943 interview: “I was trying to build up the same sort of working spirit I had seen in China where all the soldiers [of the Eighth Route Army] dedicated themselves to one idea and worked together to put that idea over. I told the boys about it again and again. I told them of the motto of the Chinese Cooperatives, Gung Ho. It means Work Together — Work in Harmony.” (Wikipedia)

However, Carlson was mistaken. The phrase is simply an abbreviation for ‘industrial co-operative.’ Still, the fact that he adapted what he thought the phrase meant for a fighting unit is interesting.

After assembling two more Raider battalions, the force was consolidated into the 1st Marine Raider Regiment, with Carlson as executive officer. His 2nd Battalion was returned to ‘normal’ structure, but his ideas were adopted first by the Regiment, and then the whole Corps.

Carlson's 3-man fire team and 10-man squad organizations were adopted, first by the Raiders and then by the entire Marine Corps. Edson contributed the concept of a highly trained, lightly equipped force using conventional tactics to accomplish special missions or to fill in for a line battalion. (Wikipedia)

It’s the last point of that paragraph which hints at the ultimate fate of the Raiders experiment.

The reality was that in the Pacific war, there were relatively few opportunities for true Commando-style raids. Increasingly, landings were done in force, using amphibious tractors, and what was needed were more line Marine units. Moreover, the lightly-armed, specialized Raiders were disliked as an ‘elite within an elite’2 by some senior officers, and the 1st was eventually redesignated the 4th Marine Regiment and brought back in line with standard force organisation.3

Enter Robin Dunbar

Though the Raiders themselves were relatively short lived, their experimentation had a lasting effect, as well as being correlated in other areas. The 3-person structure crops up over and over—in terrorist cells4 and, subsequently, in the standard force organisation of most NATO militaries, where the smallest atomic unit is typically a fire team of 3-4 personnel.

This perhaps makes sense for a reason—Dunbar’s number. Though the methodology and conclusions have been disputed, I think the theory behind it has become so widespread because people can look around them and see some correlation between real life and theory.

The reason you will probably have heard of it is the idea that you can at most maintain about 150 meaningful relationships.

However, the model is stratified, and was developed by further research. Although most people will think of the smallest ‘circle’ of relationships as 5, there’s actually an inner one, 1.5.5 If we round this up to 2, we can see that one person with two close relationships adds to 3.6

For completeness, here are the layers:7

1. 1.5: Intimates/Inner sanctum

2. 5: Loved ones

3. 15: Good friends

4. 50: Friends

5. 150: Meaninful contacts

6. 500: Acquaintances

The takeaway is this: you don’t have to deliberately always structure around these numbers,8 but it maybe pays to be aware of them. When your teams break, consider if it’s because of crossing a threshold, or due to other external factors.

Finally, always remember that if in doubt, 3-4 motivated people can almost always get the job done.