Microsoft Copilot saved people two hours a week on email — so why didn't that extra time show up anywhere else in their work?

This explores why the time Copilot saved on email didn't show up as a change in the rest of people's work. The clearest evidence comes from Microsoft's six-month randomized trial across 66 firms. Users cut about two hours a week from email, yet the amount and type of their other work stayed the same Does Copilot change what knowledge workers actually do?. The authors point to coordination. Email is something you do alone, so one person with a new tool can speed it up right away. Most other knowledge work is tied up with teammates through meetings, handoffs, shared documents, and agreed habits. That work can't be reorganized until the group changes how it works together, and in that window the group hadn't. One person's saved time can't reshape a process that belongs to the whole team.

The next question is where those two hours went. Other studies suggest they may never have been free in the first place. One line of research finds that AI often doesn't reduce total time on a task. It shifts time away from doing the work and toward writing prompts, reading what the AI produced, and checking it Does AI really save time, or just change how we spend it?. Even if email itself takes less time, some of that time gets spent supervising the AI. Measuring time-on-task misses that hidden cost.

A Berkeley Haas ethnography found the opposite of the Microsoft result. Rather than leaving work unchanged, AI made it more intense: people took on a wider range of tasks, natural stopping points disappeared, and they ran more threads in parallel Does generative AI actually save workers time or intensify it?. The two findings fit together if you look at what each one measured. A trial that counts tasks by type can show 'no change' while the pace of work and the use of breaks change in ways it doesn't record. Freed time may have been absorbed into a faster rhythm instead of into new categories of work.

There's also a social reason the gains stay hidden. In Anthropic's interview study, most workers said AI saved them time, but about 70% mentioned the stigma of using AI at work Why do workers hide productivity gains from AI use?. If people keep their savings quiet, managers and teams never redesign work around them, which is the same coordination problem seen from the inside. Email may also be a deceptively easy place to save time. Recipients trust unlabeled AI-written emails as much as human ones Do readers trust unlabeled AI-written messages as much as human ones?. But over half of desk workers admit to sending low-substance AI 'workslop', and each recipient loses hours a month dealing with it How much do desk workers actually suffer from AI workslop?. Time saved by the sender can turn into time lost by the reader.

The surprising lesson is that individual productivity tools mostly speed up individual tasks. Changing the overall shape of work requires teams to adopt the tools together and agree on new ways of working. Until that happens, saved minutes tend to be absorbed by AI supervision, a faster pace, quiet concealment, or someone else's inbox. The corpus doesn't yet include a follow-up showing what happens after teams do coordinate around these tools. That is the missing piece.