From Leviathan to AI agents

Yesterday, writing about leisure and creativity, I quoted a line from Hobbes’s Leviathan:

Leisure is the mother of philosophy.

But it actually has a second half:

Commonwealth is the mother of peace and leisure.

Leisure is the mother of philosophy, and the commonwealth is the mother of peace and leisure.

I agree with the second half too.

Because “leisure” seems like a very personal thing: having time to walk, read, daydream, think about questions with no immediate use.

But a person being able to have that kind of leisure actually depends on many things we hardly ever notice.

I can walk down the street alone without having to watch out for someone attacking me at any moment; I can buy something made by a stranger; I can put money in a bank; I can rent a room; I can trust that most contracts have some force; and if something goes wrong, there are at least places like the police, courts and regulators.

When these things work normally, it’s easy to treat them like air.

What Hobbes was discussing in 1651 was exactly this foundation.

What is Leviathan actually trying to solve?

Hobbes ran a very extreme thought experiment:

if there were no government, no laws, no courts, and no common power that could ultimately enforce rules, how would a group of people live?

His answer wasn’t simply “human nature is evil.”

He thought conflict between people could come from competition, but also from distrust and worries about safety. One person building up their strength may only be protecting themselves; another, seeing it, becomes more wary because of it.

Neither meant to be the bad guy.

But I can’t be sure you won’t hurt me, and you can’t be sure I won’t hurt you.

So each is perfectly reasonable from where they stand, yet together they may produce a result nobody likes.

That’s also what I find especially interesting about Leviathan:

many systems go wrong without needing any bad people inside them.

As long as there are several parties, each with their own interests, holding different information and unable to fully trust each other, coordination is already a problem.

Following this question outward, you find it showing up in many completely different places.

Between states: there’s no real “world government”

Inside a country there are at least a government, courts and police.

But between countries there’s no world government with real final power to enforce.

So international relations in some ways look a lot like the state of nature Hobbes described.

One country adds weapons and says it’s only for security.

Its neighbor, seeing this, can’t be fully sure those weapons won’t be pointed at it later, so it adds defenses too.

The first country, seeing its neighbor arm, becomes more convinced its worries were right.

Both can sincerely say they’re defending themselves, and end up less safe together.

International relations today calls this structure the security dilemma.

International law, military hotlines, verification mechanisms, arms-control treaties and so on are all, to some degree, dealing with the same problem:

when there’s no absolutely trustworthy final authority, how do you lower each other’s uncertainty?

In companies: a group of good people can absolutely mess up a project

Shrink the scale to a company, and you can see something similar.

• Sales wants to land the client.
• Engineering doesn’t want to leave huge technical debt.
• Design doesn’t want the experience sacrificed.
• Legal needs to control risk.
• The PM wants things to ship on time.

None of these goals is a problem in itself.

But if everyone sees different information, the goals conflict, and there are no clear decision rights, shared facts, boundaries of responsibility or ways to resolve conflict, a project can still get messier and messier.

• Sales promises early, for the sake of its own responsibilities;
• engineering then starts estimating defensively;
• the PM keeps cutting scope to deliver;
• design trusts less and less that a last-minute promise really is “just this once.”

Each action, taken on its own, can be explained.

Combined, they may turn into a very bad working relationship.

So organizational design isn’t just “finding a group of excellent people.”

Excellent people need good coordination mechanisms too.

Markets: we cooperate with strangers every day

In ordinary life, I think this is even more obvious.

Why do I dare pay first when buying a computer online from a complete stranger?

Usually not because I’ve researched the seller’s character.

Behind it there are payment providers, refund mechanisms, platform reviews, consumer protection, contracts and courts.

None of these suddenly makes that stranger more trustworthy.

What they do is something else:

lower how much mutual trust a transaction requires.

I think that’s a very useful angle for understanding many modern institutions.

A system that works well doesn’t need to assume everyone is kind, honest and always rational.

It considers: even if someone makes mistakes, acts in their own interest, misjudges, or even cheats, how can the whole system keep running?

Crypto: what if you don’t trust that “king”?

With Bitcoin, this centuries-old question took on another form.

In traditional finance, we hand the ledger to central institutions like banks to maintain.

Bitcoin asked another question:

if a group of people who don’t know each other, and don’t need to trust each other, want to agree on “who owns how much money,” can they do it without relying on the same central institution?

It tries to shift part of the trust that used to sit with institutions onto cryptography, consensus mechanisms and economic incentives.

That’s why words like trustless, consensus, governance and incentives keep coming up in crypto.

But the problems that follow haven’t disappeared.

• Who decides on protocol upgrades?
• What happens when a serious vulnerability is found?
• If a DAO is attacked, should it be rolled back?
• Who holds how much governance power?

Technology didn’t make politics and power disappear.

We just arranged rules, trust and power in a different way.

AI agents: the question is back again

Looking at AI agents now, there’s a very familiar feeling.

Suppose a system has a research agent, a coding agent, a planning agent and an execution agent at the same time.

They hold different information, can call different tools, and each takes on different tasks.

Very soon you run into some very plain questions:

• Which file counts as the real source of truth?
• When two agents judge differently, which one do you listen to?
• Who can only make suggestions, and who can execute directly?
• Who can change code, spend money, send email?
• Do high-risk operations need human approval?
• When something goes wrong, how do you know who did it?

So AI systems start to have:

permissions, source of truth, approval gates, audit logs, escalation, rollback.

These are software and AI architecture, of course.

But seen from another angle, they’re also very much a small piece of institutional design:

how a group of parties with different abilities and information can cooperate without fully trusting each other, and without letting the whole system get out of control.

We even do something similar to ourselves

A single person doesn’t always have consistent goals either.

• Today’s me wants to spend; the long-term me wants to save.
• Present me doesn’t want to exercise; future me hopes to be a bit healthier.
• The me at 11 p.m. and the me who wrote the year’s plan in January are sometimes practically two strangers.

So we set ourselves budgets, calendars, deadlines, automatic transfers, habits, even deliberately not keeping certain things at home.

Not because making rules suddenly gives me unlimited willpower.

But precisely because I know:

I can’t assume that the me of every moment will make the choice I most want in the long run.

So I leave some structure in advance for my future self.

The analogy can’t be stretched forever, of course, but I really like it.

Nearly 400 years on, some questions are still the same question

Leviathan was published in 1651.

There was no internet then, no multinational corporations, no Bitcoin, and certainly nobody discussing AI agents.

375 years later, the technology we have is completely different.

But some questions change very slowly.

When several parties, each with their own goals, holding different information and unable to fully trust each other, are put together:

how do we cooperate?

• States answered with laws and public power.
• Companies answer with organization and structures of authority and responsibility.
• Markets developed contracts, prices and credit.
• Blockchains try to answer again with consensus protocols.
• AI agents now have to answer it once more.

Hobbes’s specific political answer belongs to his own time, of course, and has always been controversial.

But the question he raised still hasn’t gone out of date.

Looking back at the line from the start, it feels a little different too:

Leisure is the mother of philosophy. Commonwealth is the mother of peace and leisure.

What we actually feel each day may just be sitting down for a coffee, reading a book, thinking about some question of no particular use.

And beneath this ordinary life lies a great deal of order, rules and infrastructure we usually can’t see, and the coordination mechanisms that let countless strangers live together.

The more stably they exist, the less we may notice them.

Until one day they stop working normally.