Why a Better AI Teammate Can Mean Lower Team Output

Background

Replacing a team member with AI that can do more work seems likely to increase the team’s output. At the very least, if AI does as much work as the person it replaces, we might expect total output to stay the same.

This paper examines theoretically how replacing a human with AI that contributes more can nevertheless reduce team output.

It looks at both the displaced person’s own work and the reasons that person gives others to work harder.

Consider a project where the more effort the leader puts in, the more another member can achieve through their own ideas and effort. If rewards are linked to team output, greater leader effort makes it worthwhile for the member to work harder too. Anticipating that response, the leader puts more effort into the project.

The member therefore contributes more than their own work. Their willingness to respond gives the leader an additional reason to work hard. If AI reproduces the member’s contribution, does it also preserve this less visible source of effort?

What the paper does

The paper models a team of two: a leader who chooses effort first and a follower who observes that effort before choosing their own. Their efforts make each other more productive. Each receives a fixed share of team output and weighs the reward from effort against its personal cost.

It compares this team with one in which AI replaces the follower. In the baseline model, AI’s contribution is set at installation and does not subsequently adjust to the leader’s effort. The leader then decides how much effort to put in based on what AI will contribute. AI input supports the leader’s productivity in the same way as human input.

The analysis first asks what happens when AI exactly matches the displaced follower’s contribution. It then lets a firm choose whether to adopt AI and how much capacity to install, considering AI costs and payments to workers. Extensions examine teams with several followers and differences in leader ability.

What it finds

First, within this model, replacing a human follower with AI that contributes exactly the same amount always reduces leader effort and team output.

With a human follower, the leader anticipates that if they slack off, the follower will notice and work less too. Replacing the follower with AI whose contribution is fixed removes this response. Even if the leader slacks off, AI does not respond by doing less. One reason not to slack off therefore disappears, and the leader works less than before.

Preserving output therefore requires AI to contribute more than the displaced follower. When the two humans have the same abilities and reward shares, and swapping their contributions leaves output unchanged, what is needed is AI that contributes as much as the leader did before replacement. With AI supplying that amount, the remaining human leader chooses the former follower’s effort level. Only then does output return to its original level.

Second, replacing one person can reduce the effort of everyone who remains.

In the extension with several followers, replacing one with AI that matches their contribution reduces leader effort through the mechanism described above. The other members, whose work becomes more productive when the leader works harder, now receive a smaller return on their own effort and work less too. Thus, replacing just one human follower with AI that contributes exactly as much can reduce the effort of every other human in the team. The extension allowing differences in leader ability also finds that replacing a follower with AI that matches their contribution causes larger output losses, both in absolute and percentage terms, in teams with a more able leader.

Third, AI adoption can raise a firm’s profit even when team output falls.

AI can contribute more than the displaced worker yet fail to offset the reduction in leader effort, leaving the team with lower output. Even so, the firm’s profit rises if savings on the follower’s pay exceed the AI cost and lost revenue.

This distinction also shapes what happens as AI becomes cheaper. Under some conditions, the firm adopts before AI contributes enough to preserve output, causing output to drop at adoption. Further cost reductions lead the firm to install more AI capacity. The larger AI contribution also encourages greater leader effort, eventually restoring output.

Why it matters

When evaluating AI adoption, comparing the work of AI with that of the person it replaces is not enough; the responses of the people who remain also matter. Matching one person’s work may leave behind the effort that their presence elicited from colleagues. This is particularly relevant when a leader acts first and other members respond afterward.

An apparent solution is simply to program AI to work harder when the leader works harder, and to do less when the leader slacks off. If that rule is reliably upheld, it can reproduce the human response and give the leader a reason to work harder. If settings can be changed later, however, those controlling AI must also have a reason to keep the rule in place.

Suppose installed AI capacity can be used at no additional cost. A leader who controls AI would prefer to use it fully regardless of their own effort. They would therefore disable the rule linking AI’s contribution to their effort and set AI to run at full capacity all the time. A firm might promise conditional use to motivate the leader, but once the leader has finished working, it too would prefer full use. If the leader anticipates this override, the promise loses its incentive effect.

A human follower chooses effort according to their own rewards and costs. Because the leader and firm cannot simply dictate that response, the leader must work harder to make greater follower effort worthwhile. Having someone who decides for themselves how hard to work can itself encourage effort from others.

This theoretical study identifies the conditions under which the mechanism operates. The decline in team output does not, by itself, mean that replacing workers with AI is a loss for society: working less also eases the burden on those who remain. The central point is that evaluating a person’s contribution requires looking at both what they produce themselves and the effort their presence encourages others to supply.

This research was supported by JSPS KAKENHI Grant Number 24K04912.

About the paper

• Title: The Hidden Value of Human Followers
• Author: Keisuke Hattori
• Year: 2026
• Publication status: Working paper, available on EconStor
• Full paper: https://hdl.handle.net/10419/344041