Engineering has primarily been about making things that operate within relatively reliable rules.

Engineers build bridges, software, vehicles, and so on.

Let's assume, just as a hypothetical, that AI can largely automate an engineer's work end to end.

What remains?

One way to answer this is to go back to Ancient China.

There were few engineers in the modern sense.

But there was already a much more fundamental problem.

How do you take millions of people, each occupied with their own private affairs, and make them act as part of a larger system?

A farmer thinks about his farm.

A merchant thinks about his trade.

An official thinks about his position.

A family thinks about its own survival.

Nobody naturally wakes up thinking about the survival of the state or the prosperity of people centuries from now.

Yet those things still have to happen.

This was one of the problems chinese ministers and the Legalists of the time were dealing with.

The emperor might see the larger picture.

But he could be erratic.

And his successor could be worse.

You cannot build a civilization on the assumption that the person at the top will always be wise.

So you need institutions that take people who are primarily concerned with their own affairs and make their actions contribute to something larger.

This is where law and bureaucracy become powerful.

People are given roles.

They make commitments.

Their performance is observed.

Their claims are compared with their results.

This is where xing-ming becomes interesting.

You say what you will do.

You are given responsibility.

Then the system checks what you actually did.

Reward and punishment connect the two.

But this is only the easy part.

The really interesting part of law begins when the system doesn't work as specified.

Someone violates the rule.

Someone breaks the contract.

Someone causes damage.

Someone delivers less than promised.

Two groups have legitimate but conflicting claims.

An accident happens.

An agent acts outside its authority.

Now somebody has to determine what actually happened.

Who was responsible?

Was there a violation?

Was there an exception?

What evidence matters?

What should the consequence be?

Who should pay?

How much?

Should the person be punished, or should the victim be compensated?

What if punishment creates a second-order problem?

What if two different groups are affected differently?

What if the rule itself produced a bad outcome?

That is law.

It is the machinery for dealing with reality after reality departs from the specification.

And that machinery is one of the foundations of large-scale civilization.

Capitalism does something similar.

It takes millions of people pursuing their own interests and connects them through prices, contracts, property, wages, profits, and competition.

The shopkeeper thinks about the shop.

The engineer thinks about the salary.

The investor thinks about the return.

Yet those private actions become part of a much larger system.

Institutions convert individual attention into collective capability.

Now AI changes the system again.

The participants will no longer be only people.

They will be people + agents.

And agents may eventually be able to spend money, access information, deploy software, negotiate, transact, hire other agents, and make decisions without a human watching every step.

So we will need rules for them too.

What is an agent authorized to do?

Who delegated that authority?

What happens when it exceeds its mandate?

Who bears the cost?

Who pays when it causes damage?

How do we reconstruct what happened?

How do we determine whether the agent, its owner, its designer, or some other party is responsible?

How do we handle conflicts between different humans, organizations, and agents?

And when the existing rules produce bad outcomes, who changes them?

This is why I think the future of work may look surprisingly legal.

AI could make the production of artifacts extremely cheap.

Code.

Documents.

Designs.

Analysis.

Plans.

But once production is cheap, the hard part becomes everything surrounding production:

setting the rules, assigning authority, interpreting the rules, detecting violations, resolving exceptions, allocating responsibility, imposing consequences, paying for damage, balancing competing interests, and changing the rules when reality exposes their flaws.

That is a very different job from simply building the thing.

The future engineer may therefore look less like a person sitting at a computer producing an artifact, and more like a person governing a system of humans and machines.

And the system will become increasingly powerful.

So perhaps the deepest effect of AI is not that machines will replace engineers.

It is that, for the first time, we may have to build institutions for a world containing vast numbers of non-human actors capable of acting on their own.

The old problem was:

How do you get millions of people, each pursuing their own interests, to contribute to something larger than themselves?

The new problem is:

How do you get millions of people and millions of agents to do the same?

That may be the real job of the future.

Practical epilogue

I've started experimenting with this idea in a small open-source project called AgentLaws:

https://github.com/shrsv/AgentLaws

The basic idea is simple.

Instead of putting everything that governs an agent into one giant system prompt, treat those instructions as a lawbook.

You have chapters, sections and individual laws.

The laws are versioned.

They have stable identities and citations.

An agent can say:

Decision: Reject Laws: 2.5.1, 2.5.3

And you can trace those citations back to the actual rule, its source, its Git revision, and its history.

This also gives humans something much better to work with.

The people governing the agents can discuss the rules, amend them, review changes through normal Git workflows, and keep a record of how the governance system evolved.

For multi-agent systems, different agents can receive the laws relevant to their roles.

This is still very primitive.

I'm not trying to solve agent governance in one library.

I'm interested in what happens when we stop thinking of agent instructions as prompts and start thinking of them as institutional law.

Because if agents are going to become participants in our systems, eventually we will need more than better agents.

We will need a civilization that knows how to govern them.