Why 4 AI Tools Is Worse Than 3

The BCG "brain fry" finding, and what it reveals about the coordination layer.

In March 2026, a BCG research team led by Julie Bedard published a study of 1,488 U.S. workers in the Harvard Business Review. The finding was counterintuitive and specific.

Productivity climbs when employees use one, two, or three AI tools. It plummets at four or more.

Not because the fourth tool is broken. Because the human absorbing all four ran out of coordination capacity. The researchers named the condition "AI brain fry" — cognitive overload caused by the mental effort required to monitor, evaluate, correct, and manage AI-generated outputs beyond the brain's processing capacity.

The physiological signature: a buzzing sensation, mental fog, difficulty focusing, slower decision-making, headaches. The operational signature: 14% more mental effort, 12% more mental fatigue, 19% more information overload.

Most organizations read this finding and conclude: "we should use fewer AI tools." That is the wrong conclusion.

The Right Read

The finding is not about tool count. It is about what the tools are asking the human to do.

When an AI tool takes over a task — writes the report, drafts the email, runs the analysis — it reduces the human's coordination load. When an AI tool adds oversight work — another output to review, another dashboard to watch, another summary to verify — it increases the coordination load.

The BCG study separated these two modes. Replacing repetitive human tasks with AI reduced burnout scores by 15%. Adding AI oversight on top of existing human work increased burnout.

The tool count correlation is a proxy. The real variable is whether each tool is absorbing work or manufacturing it.

The Structural Problem

This is why the AI-productivity debate keeps producing the wrong prescriptions.

Goldman Sachs says AI saves workers up to an hour a day. Morgan Stanley projects $920 billion in annual savings for S&P 500 companies. McKinsey's 2026 State of AI report finds that only 6% of companies are capturing meaningful EBIT impact from their AI investment.

Those three numbers are not in conflict. They describe the same situation from different angles.

The companies seeing savings are the ones where AI tools absorbed coordination work. The companies seeing the BCG cognitive-overload signature are the ones where AI tools were bolted on top of existing work without restructuring. The 6% of companies capturing EBIT are the ones who did the structural redesign that let the tools produce outcomes rather than more monitoring burden.

The technology is the same in all three groups. What differs is whether the organization's operating model allowed the technology to own work, or required the human to keep owning it.

What Coordination Work Actually Is

It helps to name what work is involved. Coordination work is the invisible scaffolding that makes business processes move: status updates, reminders, reconciliations, handoffs, meeting prep, information lookups, decision escalations, report preparation. A 2021 APQC study measured this at an average of ten hours per knowledge worker per week. Asana's 2023 data puts 58% of the typical workday in this category.

Coordination work is what managers do between decisions. It is what executives do before speaking to the board. It is what engineers do between writing code.

It is the work that scales with headcount, not with output. Every additional person in an organization adds coordination load on everyone else, not just themselves.

When AI tools are introduced without changing the operating model — when they land on top of the existing coordination structure — they do not reduce this work. They multiply it. Every tool produces its own outputs that require monitoring, its own dashboards that require review, its own verification that requires human attention. Four tools produce four overhead loads. Six tools produce six.

The BCG curve is the sound of that multiplication hitting the human's capacity ceiling.

The Alternative

The alternative is not "use fewer tools." The alternative is to restructure the work so the tools own coordination rather than produce more of it.

That is what agentification refers to. The systematic redesign of organizational work so autonomous agents — not humans — own coordination, execution scaffolding, and operational consistency.

In an agentified organization, the Status Agent prepares the weekly status, so no one is preparing it. The Decision Agent routes escalations, so no one is tracking them. The Governance Agent maintains diligence-readiness, so no one is assembling it. Each tool absorbs load rather than producing it.

This is why agentification increases operating leverage. Not because the tools are faster at tasks, but because there is less scaffolding for anyone to maintain.

The Finding, Restated

BCG's study can be read two ways. The surface reading is: "too many AI tools cause burnout." The structural reading is: "AI tools that don't own the coordination they create become coordination themselves."

The second reading explains both the BCG data and the McKinsey data. It is the same pattern viewed at two resolution levels.

If your organization is in the BCG curve — the population where adding tools makes things worse — the fix is not "subtract tools." The fix is to redesign the work so each tool owns what it was supposed to own.

That is a structural change, not a tooling change. Most organizations have not yet made it.

Sources: Bedard, J. et al. "When Using AI Leads to 'Brain Fry.'" Harvard Business Review, March 2026. McKinsey. State of AI 2026. APQC. Knowledge Worker Productivity Study, 2021. Asana. Anatomy of Work Report, 2023.