When Answers Arrive Faster, Does Work Get Easier?

Imagine turning a project meeting into a document. What would have taken half a day produces three versions in minutes with AI, complete with orderly headings and readable paragraphs. It looks as though all you need to do is choose one.

Then you put them side by side. One explains the background thoroughly, another makes its conclusion prominent, and a third looks useful for assigning work. You still need to understand them, check the facts and explain to colleagues why one should stay. The keyboard gets a break. The brain may still be on the clock.

A few more versions—and someone has to read them

“It won't take long. Let's see them all.” It is an understandable thought. Three versions used to mean considerably more writing, so we would at least ask why we needed three. Now we can add the request almost as an afterthought. Each document takes less time to produce, but more documents may be waiting for someone to read.

More possibilities can be useful, particularly before the direction is clear. Each candidate, though, adds something to understand, verify and compare. A printer that produces dozens more pages a minute does not help a meeting reach dozens more agreements.

If the documents offer different approaches and explain their resources, deadlines and costs, reading them may lead to a choice. If they merely swap adjectives, we may reach the end before discovering there were never three distinct options. There are more drafts, and the original decision is still waiting.

AI can summarize differences and filter candidates first. But comparison needs criteria. Does speed matter more here, or lower risk? Do we need a full account of the background, or enough for a particular reader to make a decision quickly? Without clear criteria, sending the drafts back to the model for a verdict is still asking it to guess for us.

Of course, faster writing can save time when the goal is clear, the number of drafts stays the same and the important claims are easy to check. Generative AI at Work found that the effects of AI assistance varied with experience and skill in a customer-support setting, with less experienced workers benefiting more.^[1]

That study makes me want to ask where the difficulty was to begin with. If it was mainly putting words together, AI can help directly. If we have not yet worked out what decision to make, extra drafts may offer less relief. The added review work is my reasoning about this hypothetical task, not a finding of that customer-support study.

What is this document going to be used for?

The three drafts may be trying to do different jobs.

One person wants absent colleagues to understand the discussion. Another wants to arrange follow-up work. A third needs to decide whether the project should continue. Absent colleagues need background and disagreements; follow-up work needs decisions, owners and unresolved items. A decision about continuing needs evidence of progress and conditions that remain unmet. The same recording could reasonably produce three different good documents.

But if the request was only “make it professional,” the model has to guess which purpose to serve. Each version may be well structured, yet none feels quite right. Better wording may not be what is missing. We may still be unsure who is supposed to use the document, and for what.

Without a clear purpose, the model guesses, and the reviewer has to correct those guesses. Thinking that belonged in the original request has been postponed until review. Establish which decision the document should support, and it becomes easier to explain which background matters and which conditions a conclusion must retain.

The request should also say what the meeting settled and what remains undecided. A column for a task owner cannot supply someone the meeting never appointed. Layout and wording can be left to the model; every comma does not need instructions. Once the choices that affect meaning and consequences are clear, we have room to delegate the rest. Otherwise, we put down the pen only to approve every detail of what gets written.

Why keep the less impressive sentence?

Suppose this document will be used to discuss expanding a pilot. One sentence deserves a close look: “The pilot is complete; the effects of wider rollout remain to be seen.” One version changes it to “We have achieved a comprehensive breakthrough.”

The new sentence sounds stronger, but merges completion of a pilot with evidence that expansion works. We sometimes call this polishing, as though the facts should help the prose look its best.

Readers may infer that the conditions for expansion have been met. The discussion moves from “What do we still need to learn?” to “When do we start?” A sentence changes their understanding of the evidence, which may then affect resources and commitments. Keeping or removing it is no longer only a matter of style.

What I call taste includes noticing this change and explaining why the plainer sentence fits. “It feels wrong” leaves others unsure how to revise next time. Explaining that a completed pilot does not establish the success of a wider rollout gives everyone something to check. Style is negotiable; missing evidence cannot be supplied by better wording.

Even verified facts may leave more than one reasonable choice. The pilot could expand quickly or stay small for a while, depending on deadlines, whether the decision can be reversed, and the costs people can absorb. Change those conditions, and the same facts may support another choice. When I read “On balance, we recommend…,” I want to look again: what was considered, and who will pay for anything that was left out?

Software can check fields, and AI can compare drafts and find evidence. A person need not repeat every step. Disputed choices can go to someone with the relevant knowledge and authority, but that person needs to see the evidence. Adding an “approve” button at the end does not save the effort of making a judgment.

In my course, The Art of Human–AI Collaboration in the AI Era, I use matter, energy and information to discuss how technology changes work. It is a teaching framework with overlapping stages, not a law of history.

This document brings those connections into view: understanding its purpose, explaining the assignment, choosing words and making a decision affect one another. An unstated purpose can mean several rounds of revision; a missing condition can change someone else's choice. These are problems of preparing and using information. Physical skills, care, relationships and coordination on the ground deserve their own discussion.

Faster answers can certainly mean less typing. Making the work easier sometimes starts with explaining what the document is for. Sometimes it simply means keeping an unexciting sentence: “The effects of wider rollout remain to be seen.”