What AI Left Behind
Gensler’s Research Institute published its 2026 Global Workplace Survey last week. Sixteen thousand workers. Dozens of countries. A lot of questions about how AI is changing the nature of work.
The finding that didn’t make the headlines: workers most deeply embedded in AI tools are the most connected to their colleagues. Most collaborative. Most interpersonally engaged. Not most isolated.
The expected narrative says AI separates people from each other. Automates human contact out of the workday. Makes work a solitary, human-machine interaction.
The data says the opposite.
What automation actually removes
The explanation isn’t complicated, once you see it.
AI handles the parts of work that don’t require other people.
The first draft. The data pull. The summary of last quarter’s numbers. The scheduling back-and-forth. The status update. The research synthesis. All of it — solo work. Heads-down, machine-compatible, interaction-optional.
When that work moves to AI, what remains is what couldn’t move. The judgment call that depends on context only three people in the company hold. The decision that requires reading the room. The coordination overhead that shows up as “can we get everyone aligned before this ships.” The relationship maintenance that’s actually load-bearing.
AI didn’t isolate knowledge workers. It filtered out the parts of work that were already isolated and left the parts that require people.
The most AI-dependent workers are the most collaborative because they’ve pushed the solo work down to the machine and they’re spending their time on what remains. The stuff that only works with other humans in the loop.
The 14 percent
BCG published different data this year.
They studied 1,488 workers who use AI daily. By any measure, people who’ve adopted the tools. Not resistors. Not reluctant users. Daily AI users.
Fourteen percent of them are experiencing what BCG calls “brain fry” — cognitive overload. The kind that shows up as decision fatigue, inability to focus, lower output quality by end of day.
The cause isn’t their workload. It’s the supervisory overhead of AI.
Every AI output requires a decision: good enough, or fix it. Every automated task generates an artifact someone has to evaluate. Every agent produces results someone has to route to wherever they’re supposed to go. Every delegation creates a monitoring obligation.
That monitoring adds up. Fourteen percent of daily AI users are now carrying more cognitive load, not less, because the monitoring overhead of AI delegation exceeded the cognitive savings of AI assistance.
These aren’t people who adopted AI wrong. They adopted it the way they were supposed to. The missing piece isn’t their behavior — it’s the infrastructure that should be absorbing the monitoring overhead that’s currently falling on them.
Two findings. One gap.
The Gensler finding and the BCG finding seem to contradict each other. Same year. Same population — daily AI users. One group is more connected, more collaborative, less isolated. The other group is overwhelmed.
They don’t contradict. They describe two different experiences of the same structural problem.
The workers who are more collaborative — they have somewhere for AI output to go. There’s a system, or a team structure, or a workflow design that routes what AI produces into decisions, into coordination, into action. They don’t absorb the monitoring overhead themselves. The structure does.
The 14% with brain fry are the structure.
Every AI output lands on their desk. Every error returns to them for correction. Every automated task requires their sign-off because nothing decides what needs sign-off and what doesn’t. There’s no filter between AI-produced content and human-required review. No routing layer. No coordination infrastructure.
They’re experiencing what happens when you automate the tasks but not the management of what automation produces.
What this reveals about the next layer
This pattern is legible now if you know where to look.
AI made the task layer faster. Output per hour went up. First-draft time collapsed. Research that took days takes hours. The individual productivity gains are real.
What didn’t get built is the management layer above the task layer. The system for deciding which AI outputs require human review and which don’t. The infrastructure for routing what AI produces to whoever needs to act on it. The coordination layer that closes the loops AI opens, tracks the commitments AI makes on someone’s behalf, surfaces what matters versus what can resolve below their attention.
The Gensler workers who are thriving — something is doing that management job for them. Sometimes it’s a team structure. Sometimes it’s a workflow that enforces routing. Sometimes it’s an organization that built coordination practices ahead of AI adoption.
The 14% don’t have that. They became it.
That’s what AI left behind when it automated the task layer. Not the tasks that require creativity or judgment — though those too. It left behind the coordination overhead that comes with managing an increasingly capable, increasingly active system that doesn’t know what to do with what it produces.
The question this raises
Not: “How do we help people use AI better?”
That’s the individual layer. The training programs, the prompt engineering, the personal workflow optimization. Necessary, insufficient.
The real question: “What absorbs the management overhead of AI operating at scale inside someone’s work?”
For the Gensler workers who are more collaborative, something does. For the 14%, nothing does.
The difference between those two groups isn’t effort or skill or tool selection.
It’s whether there’s infrastructure between the AI and the human — or whether the human is the infrastructure.
That’s the gap. It’s not a technology gap. It’s not a skill gap.
It’s a coordination layer that hasn’t been built yet.
