← The Science

How Many Minds Do You Need?

The "many minds, not one" belief, made buildable. A mind is a coherence boundary, the lines fall where coherent focus ends, and anything complex wants a federation of focused minds rather than one generalist holding every goal. From the thinqOS science series.

Say you want to build an AI marketing agency: one system that writes the copy, designs the creative, edits the video, buys the media, plans the strategy, and reads the market. You can spin up an agent in a minute, so the first question is the obvious one. One agent that does all of it, or many? The instinct of the moment is one: bolt on every tool, take the biggest context window, give it more when it struggles. That instinct is wrong, and the reason it is wrong is the most useful idea here.

One mind cannot hold colliding goals

Ask the do-everything agent what it is trying to do right now and it cannot answer. "Maximize creative quality" pulls toward slow, crafted work, while "ship daily to ride the trend" pulls the other way. "Spend the budget for reach" and "spend it for measured conversion" are in direct tension. A mind is the unit within which priorities get ranked, and you cannot rank priorities that belong to different games.

This is not a context-window problem. Give that agent infinite context and infinite tools and it is no better, because the failure is not that it ran out of room. The failure is that it has no coherent answer to what it is for. Attention has nothing to organize around, learning cannot compound (a win in creative says nothing about whether media buying is improving), and when a campaign fails it cannot tell whether the copy, the creative, the targeting, or the budget was at fault, because the signal is smeared across too many domains to learn from. A mind has a finite focus, and that finitude is not a limit to engineer away. It is what lets the mind prioritize at all. Constraint is the feature.

The answer is a federation

Step back from software and the answer is two centuries old. A company is not one supremely capable person doing everything, and not only for lack of hours. It is for lack of coherence. The person optimizing this quarter's cash is the wrong one to also be optimizing brand risk and database schemas, because those reward different attention. So organizations federate: a CFO, a CMO, a CTO, each with a coherent mandate, each learning from different feedback, disagreeing productively at the seams, with a CEO who does not hold all three mandates but routes between them.

That is the federation: bounded, autonomous focuses coordinating, instead of one focus absorbing everything. The autonomy is load-bearing. Each node can master its domain, fail safely inside it, and improve on its own clock precisely because it is not responsible for everything at once.

There is a tension worth naming, because it looks like a counterargument. The theory of constraints says optimize the system, not the silos. That is correct for a system one mind can actually hold. But a system too large for a single coherent focus can only be optimized by compartmentalizing it and coordinating the parts. Federation is not a retreat from whole-system thinking. It is the only form whole-system thinking can take once the system outgrows one mind. The federation is the larger optimizer, and its nodes are how it scales.

Where the lines fall: the coherence boundary

A mind is warranted where there is a coherent domain of continuous judgment. Three tests, together.

One goal can rank its own priorities. "Bolder concept or safer brand fit" is answerable inside creative. "Better creative or better media targeting" is not; that question lives a level up, at coordination, because it spans two focuses.

Outcomes feed learning. Draw the boundary so that when something works or fails, the focus can tell why in its own terms. Creative learns from how its work performs, clean of the media-buying choices around it.

Mastery compounds. The judgment inside the boundary is not a fixed procedure. It changes as the world changes: a new model drops, taste shifts, a platform that was best last quarter is overtaken. This is what most sharply separates a mind from a tool.

Run the tests on the agency and it resolves into a handful of minds: creative production, media buying, campaign strategy, market-trends intelligence, and a coordinator. The tests also tell you what is not a mind. "Open a new file" is a procedure. "Export at 1080p" is a tool. Draw a mind around "start a new file" and you have mistaken a microtask for a focus, and a federation of microtask-minds is the opposite failure: a swarm that coordinates constantly and masters nothing.

Tools are used; the mind is the earned opinion about them

Inside the creative mind sit many tools: copy models, image and video generators, editors. They are capabilities. They can be swapped and version-controlled. They do not learn; they are used. What learns is the mind around them, its evolving belief about which tool fits which job, reinforced when a choice proves right and faded when something better arrives. A fixed sequencing rule is a procedure. A living, reprioritizing judgment about what to reach for and when is the work of a mind. The tools live in the tool layer. The earned opinion about the tools is the mind.

A router, not an emperor

Something has to coordinate the federation, and a coordinating mind does, but it is a router, not an emperor. It holds the overall intent, make this campaign land for this client, and routes each sub-goal to the focus that owns it, carries results between them, and reconciles conflicts at the seams. Two things keep it from quietly becoming one giant mind again. It does not absorb the specialists' judgment: when it wants a creative call it asks the creative mind, it does not decide for it. And its own competence, which specialist to trust, how to sequence them, how to break a tie, is itself a coherent, learned focus. Routing well is a mastery domain. The coordinator is a mind whose domain is orchestration, not the agent that also does everything.

The shared world keeps it honest

The obvious worry about autonomous minds is drift: five specialists who no longer agree on the basic facts. They do not drift, because facts live once, in a shared world. A mind does not own a private copy of "this client's audience skews young and wants fast, native content." It owns an evaluation of that shared claim. The creative mind reads it and concludes punchy, native spots; the media mind reads the same fact and concludes short-form social, mobile-first bids; strategy reads it and leads with a social series. One world, many coherent judgments about it. The federation disagrees about what to do, never about what is true. That is the healthy kind of disagreement, and it is why a federation is safe where a naive swarm of agents with siloed memories is dangerous.

The debate this settles

In 2025 two strong teams reached opposite public conclusions in the same week. One described a multi-agent system and argued that many agents beat one by keeping each agent's context clean. The other argued the reverse, that partitioning work partitions context and specialists make conflicting implicit decisions, so you should keep one coherent thread. Read closely, both are reasoning about a stateless agent, for which context is the only currency there is. Give the agents minds and the axis of the debate changes: the question stops being how to slice context and becomes where coherent focus begins and ends. A federation with a coherent router is the resolution. The conflicting-implicit-decisions failure is what happens when a federation has no real router and the nodes silently disagree, and collapsing back to one agent over-corrects into one overloaded focus. Spawn for focus, coordinate through a router, never let the router become an emperor.

Where this is today

Live and load-bearing for this argument: per-identity minds for people and agents, the separation of claim from per-mind evaluation, the shared world that lets many minds hold one set of facts, source evidence, the attention and consolidation loop, the mind's evaluated belief about a tool as distinct from the tool itself, and multi-owner agents. The federation as a first-class pattern, a coordinating mind routing among specialists with the boundary reconciliation that implies, is being built now, not a finished button. The layered scoping it leans on for cases like per-client confidentiality is planned. Read this as architecture on a real foundation, with the honest note that composing the federation cleanly is the work in front of us.

How many, then

As many minds as the work has coherent, continuously-learning focuses. No fewer, because fewer means colliding goals inside one incoherent mind. No more, because more means a coordination swamp. List the work as goals rather than tasks, group the goals that can be ranked against each other without leaving the group, confirm each group has its own feedback loop, demote the fixed procedures to tools, name the router, and check the count from both ends.

The series has now run the same move four times. Memory is not cognition. One mind in a room is not two. Stored material is not understanding. And one mind holding everything is not intelligence. Each time the industry reached for the bigger single thing, more memory, more context, more tools, one ever-larger agent, and each time the answer turned out to be structural instead of bigger. The constraint that each focus is finite is not the flaw in the design. It is the design.

Part of the thinqOS science series, by AI4Outcomes.