The Work Doesn’t End at Send

Feature image: AI-generated illustration from the Signal Dispatch archive.

The question started with coding agents: what happens when writing code becomes easier than reviewing it? Then it widened. Imagine personal assistants writing more emails, taking more meetings, and sending more follow-ups. What happens to the people who have to respond?

An assistant can make a request cheap to compose without making the recipient’s decision cheap to make. If easier communication encourages more requests, the time saved by one person could become work distributed across everyone else. Whether the assistant helped depends on what happened to the whole exchange.

That is where Jevons paradox and the idea of doing less meet. A tool that helps us produce more activity needs some way to recognize when the useful work is already done.

Cheaper work can invite more work

In The Coal Question, William Stanley Jevons argued that using coal more efficiently could increase total consumption. Greater efficiency made more industrial uses profitable, expanding demand enough to outweigh the savings in each use.

The important part is the expansion. Making an activity cheaper changes which activities are worth attempting. Something that previously cost too much becomes reasonable to do, so the amount of activity changes along with the cost of each instance.

That offers a way to think about AI assistants. It does not establish that they will increase everyone’s workload. Coal consumption and human attention are different things, and a historical argument about one cannot settle what happens to the other. The connection is a hypothesis about what people might do when a constraint gets easier to work around.

Consider a hypothetical request for feedback on an early proposal. Writing the explanation, deciding whom to include, and finding a time to discuss it might previously have been enough effort to delay the request. An assistant could prepare all of that from a few notes. It could also follow up when nobody responds.

For the sender, that is a useful reduction in effort. For the recipients, the proposal still needs to be understood. Someone has to decide whether it deserves attention now, whether the right people are involved, and what agreeing to it would mean for work already underway.

Their assistants could help too. They might summarize the proposal or prepare a response. But if the exchange creates a new commitment, easier reading and writing have not necessarily made that commitment easier to fulfill.

The same issue applies to meetings. Imagine an assistant attending on someone’s behalf and returning a summary with suggested actions. The person may have saved the time they would have spent in the meeting. Whether they have less work afterward depends partly on what they are now expected to do.

A summary can make an obligation easier to understand. It does not, by itself, remove it.

Doing less has to survive delegation

In his Big Think interview about slow productivity, Cal Newport challenges the use of visible activity as a substitute for useful effort. His proposal includes doing fewer things at once, working at a natural pace, and concentrating on quality. Doing fewer things at once is meant to support worthwhile accomplishment over time, rather than keep every possible project moving simultaneously.

That is his argument about work. Applying it to assistants raises a further question: what happens when we delegate the activity but leave the standard of success unchanged?

An instruction to keep following up until someone replies gives an assistant a clear task. It does not tell the assistant whether another reply is worth pursuing. An instruction to find a meeting time gives it a scheduling problem. It does not ask whether the meeting is necessary.

Those are different decisions, and making the first one easy does not answer the second. If doing less matters, it has to appear somewhere in what the assistant is asked to accomplish. Otherwise, we could become better at maintaining commitments we should have reconsidered.

For that hypothetical proposal, the useful outcome might be a decision about whether to proceed. The assistant could gather the missing information and identify the person who can make that decision. Once the decision is made, it should be able to stop. Another check-in is not automatically progress.

This also puts a limit on the promise of a personal assistant. One person can delegate their follow-ups without having the authority to reduce everyone else’s obligations. In a workplace that expects every request to receive a response, an assistant cannot solve the problem through better wording alone. The expectation needs to change too.

More communication can be worthwhile

None of this makes difficulty a good filter for what deserves to happen. An important request can go unwritten because someone lacks time, finds writing difficult, or cannot work out how to explain what they need. An assistant that helps them communicate could create more messages and still make the situation better.

Nor would fewer messages necessarily mean less work. A vague request might need several clarifications. A well-prepared one could include enough context for a useful decision immediately. The longer message could save effort across the exchange.

The question is what the additional communication enables. Does it help someone resolve something that matters? Does it make a necessary commitment clearer? Or does it keep a low-priority possibility active because producing the next message has become almost effortless?

An assistant could work in either direction. It might notice that a question has already been answered, combine overlapping requests, or point out that a proposed meeting has no decision for anyone to make yet. Those are possibilities worth designing for, alongside faster drafting.

But reducing activity needs care as well. An unanswered request is not necessarily an unnecessary one. The assistant needs enough context, and appropriate authority, to tell the difference. Where it cannot, it should bring that decision back to the person responsible.

What the comparison would need to show

A useful evaluation would follow comparable tasks through to their outcomes, with and without assistant support. Drafting time would be part of it. So would the effort asked of recipients, the commitments created, and whether a useful decision was reached.

The unit of comparison matters. Looking only at a message could miss the follow-ups it triggered. Looking only at the sender could miss the work shifted to everyone else. Counting fewer meetings could miss a necessary discussion that never happened.

The standard cannot be maximum activity or minimum activity. It has to be whether the work was worth doing and whether the assistant helped finish it with less unnecessary effort. That requires judgment from the people involved; a count of generated messages cannot supply it.

There is no result from that comparison here. Jevons gives us a reason to question whether cheaper activity will stay at the same volume. Newport gives us a reason to question whether more simultaneous activity is desirable. The effect of a particular assistant still needs to be examined in the setting where people use it.

An assistant could help prepare a proposal, get it to the right person, and clarify the decision. If the decision is to leave it alone, that should be a completed task. There should be no follow-up waiting to bring it back next week.