When Coordination Is Avoidable: A Monotonicity Analysis of Organizational Tasks

Imagine you are running a massive, chaotic kitchen where hundreds of chefs are trying to cook a giant banquet. In the old days, the head chef would shout orders, check every single plate, and make sure everyone was talking to each other constantly. This "checking" is called coordination. It's like a referee blowing a whistle every time two players might bump into each other. But here's the thing: sometimes, the chefs are just chopping vegetables for their own separate salads. They don't need a referee to tell them when to chop; they just need to chop, and then mix the bowls at the end. If you have a referee watching every single chop, you waste time and energy.

In the world of science, this problem shows up in two places: in human companies (where meetings and approvals slow things down) and in computer systems (where "agents" or AI programs try to work together). Scientists have long known that some tasks need a referee, while others don't, but they didn't have a perfect rule to tell the difference. They had a list of "types" of teamwork, but it was vague. On the other side, computer scientists had a strict math rule called monotonicity. Think of monotonicity like a one-way street for information: once you add a fact to your story, you never have to take it back. If your story only grows bigger and never shrinks or changes its mind, you don't need a referee. If your story might need to be rewritten because of new information, you do need a referee to make sure everyone agrees on the changes.

This paper asks a simple but huge question: Can we use that strict computer rule to figure out which human and AI tasks actually need a referee, and which ones are just wasting time? The authors, led by Harang Ju, decided to test this by looking at thousands of real-world jobs and computer simulations. They wanted to see if we could stop paying for "referees" on tasks that don't need them, saving a massive amount of time and money.

The Great "Referee" Audit

The paper starts by building a bridge between two different worlds of thinking. On one side is Thompson's Taxonomy, a classic way of sorting tasks into three buckets:

1. Pooled: Everyone works alone and dumps their results into a big pile (like a potluck).
2. Sequential: One person finishes a step and passes it to the next (like an assembly line).
3. Reciprocal: Everyone has to talk to everyone else constantly to adjust their work (like a jazz band improvising).

On the other side is the CALM Theorem from computer science, which says: "If your task is monotonic (you never have to take back a conclusion), you don't need coordination to get the right answer."

The authors' big discovery is a "Bridge Theorem." They found that Pooled and Sequential tasks are almost always monotonic. This means if you have a potluck or an assembly line, you can often let the workers run wild without a referee, and the final result will still be correct. The only time you really need a referee is when you have Reciprocal tasks where the feedback is "retractive." This is like a teacher saying, "You wrote this essay, but now I'm telling you to delete the whole first paragraph." That's a retraction. If the feedback is just "add more details" (additive), you don't need a referee. But if the feedback says "change your mind," you need one to make sure everyone agrees on the new version.

What They Found: The "Coordination Tax"

The authors didn't just stop at theory; they went on a massive audit. They looked at 65 real enterprise workflows (like how a bank processes a loan or a hospital handles a patient) and 13,417 different job tasks from a giant database of occupations.

Here is the shocking part:

• 74% of the enterprise workflows were "monotonic." This means nearly three-quarters of the complex business processes they looked at do not need coordination to be correct. They could run without the constant meetings, approvals, and status checks.
• 42% of the individual job tasks were also monotonic. Even for a single person's job, if you break it down into steps, almost half of the time, the steps don't require a "referee" to ensure the final answer is right.

The paper calculates something called the "Coordination Tax." This is the amount of money and time organizations waste by treating every task like it needs a referee, when only the "retractive" ones actually do. Depending on how expensive the coordination is, the authors suggest that 24% to 57% of the money spent on coordination might be unnecessary for getting the job done correctly.

The AI Simulation Proof

To prove this wasn't just a guess, the authors ran a video game-style experiment with AI agents. They set up three different AI models (think of them as three different types of smart robots) and gave them ten different tasks.

• The "Coordinated" Group: The robots had a manager who told them what to do, checked their work, and fixed conflicts.
• The "Uncoordinated" Group: The robots just did their part and slapped their answers together, with no manager.

The results were clear. For the "monotonic" tasks (like writing a strategy or making marketing content), the uncoordinated robots got 100% of the answers right. They didn't need a manager! But for the "non-monotonic" tasks (like allocating a limited budget or scheduling a production line where resources run out), the uncoordinated robots failed 100% of the time. They needed the manager to fix the conflicts.

The Takeaway

The paper doesn't say that all meetings are bad or that we should fire all managers. It says that for a huge chunk of work—especially the kind where people just add their piece to the puzzle or pass a baton down a line—we are probably paying for coordination that isn't needed for correctness.

The authors are careful to say this is a "theoretical upper bound." It means that if we switched to a system where we only coordinated when absolutely necessary, we could save up to half our coordination costs. In reality, companies might still coordinate for other reasons, like making sure everyone's style matches or moving faster. But the paper proves that the idea that "coordination is always necessary" is wrong. For many tasks, the best way to work is to let the agents run free, trust that their work only grows and never shrinks, and only call in the referee when someone needs to take a step back and change their mind.

In short: Stop checking the potluck dishes before they are served. Only check the ones where someone might have to throw the whole meal away and start over.