What Is Separation of Powers? (Key Term Explained)
In the previous chapter, we looked at a third option beyond "slave" and "runaway machine" -- organizing AI by dividing roles and building in mutual checks. We called that approach AI organization (here meaning: designing AI agents around clear, separated roles with checks between them).
In that context, we introduced the term separation of powers (the principle of dividing authority between distinct bodies so each can check the others). This chapter takes a closer look at what that term actually means. It is not a difficult idea. You likely saw it in a civics class at some point. But we want to be careful about exactly which part of it matters for this series.
Q. Where does "separation of powers" come from?
It is originally a term from political science -- the study of how governments are structured.
The idea is this: divide a nation's authority into three parts, and have each part watch over the others. A political thinker in 18th-century France argued that power should never be concentrated in a single place. That idea became the foundation for most modern democracies, and it is built into their constitutions.
Q. What are the three powers, specifically?
In the case of Japan -- and in most democracies -- the three are as follows.
Legislative (the power to make laws): This belongs to the legislature (parliament or congress). It is where decisions are made about what the rules will be.
Executive (the power to carry out the laws): This is the branch that actually runs the country day to day -- the cabinet, ministries, and government agencies. It puts the decided rules into action.
Judicial (the power to judge under the law): This belongs to the courts. When a dispute arises or a rule may have been broken, the courts decide whether a violation occurred.
Put simply: make the rules, carry them out, judge against them. Three separate functions, held by three separate institutions.
Q. Why go to the trouble of splitting into three?
Because when everything is in one place, there is no one left to say stop.
Imagine a single person who writes the rules, enforces those rules themselves, and then decides whether their own actions broke those rules. No matter what they do, no one can say "that is wrong." There is no check on them at all.
Separation of powers exists precisely to prevent that situation -- a state where no one can stop the person in charge.
When authority is split three ways, if one branch overreaches or starts to go wrong, the other two can apply the brakes. If the legislative branch tries to pass a rule that goes too far, the judicial branch can rule it unconstitutional. That is the mechanism in action.
Q. What does "checks and balances" mean?
Checks and balances is the name for the relationship between those three branches: each one restrains the others and keeps the overall system in equilibrium.
"Checks" means each branch can push back against the others when they overstep. "Balances" means the result is a system that does not tip too far in any one direction.
This is the heart of separation of powers. It is not just about dividing things into three. It is about designing those three so that they actively watch each other -- and can step in when needed.
This is exactly the framework we want to borrow for this series.
Q. Separation of powers is about governments, right? Is it relevant beyond that?
Yes, it originated as a framework for national governance.
But the underlying question -- "what happens when authority concentrates in one place?" -- is not limited to governments. Any time multiple people or organizations are running something together, that question applies.
In the previous chapter, we talked about designing our relationship with AI as a structure where roles are divided and each part checks the others. Separation of powers is the foundation that thinking stands on.
What we are borrowing here is not the governmental structure itself. We are borrowing the skeleton: do not concentrate authority in one place -- divide it, and have the parts monitor each other.
Q. What do the three parts map to in this series?
To give a rough preview: the three divisions we will work with are execution (the role that acts), audit (the role that checks), and approval (the role that makes the final call).
Who moves, who checks, and who gives the final decision -- that is the role design we will build toward.
The detailed breakdown comes in later chapters. The goal of this chapter was simply to establish what "separation of powers" means as a term, and where its core idea lies.
To sum up: separation of powers means dividing a government's authority into the legislative (makes laws), executive (carries them out), and judicial (judges under them) -- with each branch holding checks and balances over the others -- so that power cannot concentrate in one place or run unchecked.
That design principle, born from political theory, is what we will connect to the question of how AI agents are organized. Now that the term is grounded, we can start building on it.