AI Saved You Time. Why Is Your Calendar Still Full?

A useful result with a limited claim

Dillon, Jaffe, Immorlica and Stanton's Shifting Work Patterns with Generative AI, a 2025 NBER working paper revised in November 2025, reports a field experiment across 66 firms and 7,137 knowledge workers. In the latter half of the six-month experiment, the users among treated workers spent about two fewer hours on email per week and worked less outside regular hours.

The authors did not detect wider changes in the quantity or composition of tasks from individual access to the tool. The distinction matters: individual time savings can be valuable even when they do not reorganise a team's work. Some authors were employed by Microsoft, whose tool was studied; the paper discloses that relationship. This is evidence from a particular deployment, not a promise of two hours back for every employee.

The calendar belongs to more than one person

An email assistant may shorten one person's drafting time. It does not automatically cancel the recurring meeting, assign ownership to an unresolved decision or persuade another team to approve a request sooner. The worker controls part of the workflow; the organisation controls other parts.

Our interpretation is that AI adoption and work redesign should be treated as linked management tasks. A tool can create capacity, but a team needs to decide whether that capacity will go toward deeper work, better service, recovery time or more volume. Leaving the decision implicit can make a helpful tool feel like another demand.

Run a practical time-savings review

Choose one repeated activity, such as weekly reporting or routine email preparation. For two weeks, note the time spent producing the work, checking it, waiting for input and correcting it. Use only information you are permitted to record, and keep confidential correspondence out of public examples.

Then identify the next constraint. If drafting is faster but the work waits for approval, try clearer approval criteria or an agreed review window. If checking consumes the saving, improve the inputs or reduce the tool's scope. The aim is to locate a change the team can actually make.

This small exercise is a proposed working practice, not an experimental design that can establish a general effect. Its value is in replacing a vague impression with a specific discussion about where time goes.

Protect the saving from becoming invisible extra work

Agree what a better result looks like before raising output targets. More messages are not necessarily better communication, and more reports are not necessarily better decisions. A team can prefer fewer, clearer updates while using AI to improve their accuracy and preparation.

For remote staff, boundaries deserve explicit discussion. If faster daytime work merely creates more evening requests, a measured improvement in one task can coexist with a worse working week. Ask when a response is expected, which decisions can wait and how urgent issues are escalated.

The research invites a more precise question than whether everyone should use AI. Which workflow should change, what outcome should improve, and how will the people doing the work notice the benefit?

A conversation to bring to your next team meeting

Bring one example of a task the tool helped and one part of the process it did not change. Propose a single adjustment, such as replacing a status meeting with a short decision log or reserving a review slot. Evaluate whether the adjustment helped rather than assuming a new tool has solved the whole process.

AI time savings do not need a dramatic headline to matter. Less evening email, clearer work and fewer avoidable handoffs are useful outcomes. The calendar stays full until someone gives the saved time a purpose.