Note · AI and work · 2026-09-20 · 4 min read
When producing more becomes the problem
AI makes it easier to produce more, but that can simply shift work onto the people receiving it. This article looks at how to keep AI-generated work useful, concise and worth someone else’s attention.
By Dr Zhitao Xiong · in a personal capacity
There is no shortage of things to write about in AI. There is a harder question: which of them deserves another few minutes of your attention?
AI can turn a few thoughts into a polished report, a meeting into a list of actions, or a vague request into a substantial proposal. That is useful.
But will anyone read it?
Sometimes I open one of these documents and think, “Hmm, too long, with traces of AI everywhere. Maybe later.”
Producing something has become cheap. Making it worth passing on still takes work.
AI does not necessarily remove work. Sometimes it simply moves the work downstream.
You are not AI’s “DoorDash”. You still own what you send.
Imagine a manager asking AI to produce a strategy document and forwarding it to the team with “Thoughts?” It takes minutes to generate. Several colleagues then spend an afternoon checking the claims, untangling the recommendations and working out what the manager actually wants.
My wife used to receive 50-page “strategy documents” from her manager that appeared not to have been proofread. The AI had produced the document; her manager had mostly delivered it.
The sender’s time saving can become the recipient’s workload.
This is why, even before the AI hype, I hated receiving voice messages on WhatsApp or WeChat. A three-minute recording that I have to listen to from beginning to end could often have conveyed its actual point in a few lines of text.
The same pattern again, now with AI as the excuse.
Researchers at BetterUp Labs and Stanford’s Social Media Lab call a version of this problem “workslop”. In their September 2025 survey of 1,150 full-time US desk workers, 40 per cent reported receiving it in the previous month, with roughly two hours spent resolving each incident on average. These are self-reported estimates, but they highlight a cost that individual time-saving claims can miss.
Productivity measured only at the point of generation can therefore be deeply misleading.
A document can be accurate, articulate and completely unnecessary. Checking facts matters. So does deciding whether anyone needs the document in the first place.
AI has dramatically increased our capacity to produce. It has not increased everyone else’s capacity to consume.
That changes what is scarce.
Content is becoming abundant. Attention, judgement and organisational bandwidth are not.
So I would start with six habits.
Write your own thoughts first. Use AI as a writing partner.
Whether it is a Jira update, a report or a brief for executives, start by organising your own thoughts into bullet points. Ask AI to polish the wording, challenge your reasoning and identify missing detail. Check any additions against the evidence, then read every word before sending.
Before generating, name the decision.
Before asking AI to create anything, finish this sentence:
“After receiving this, this person should be able to decide or do…”
If you cannot finish that sentence, clarify the purpose first.
Sometimes a paragraph will do. Sometimes a conversation will do. Sometimes there is simply nothing material to update.
The question is not whether AI can produce the document. It is whether the document needs to exist.
Before sending, do your part.
Read the work. Check the important claims. Put your recommendation and request at the top.
“I recommend testing option B. I checked the costs. Demand remains uncertain. I need your approval for a small trial.”
Keep the supporting detail available.
If you want help with an unfinished idea, say what is unfinished and which part needs input. Do not make the recipient reverse-engineer your thinking.
When reviewing, challenge an assumption.
Ask what could change the recommendation and examine the evidence behind it.
For a market proposal, that might be demand or delivery cost.
Give reviewers time and authority to return work that is not ready, including work from someone more senior. AI makes producing another draft cheap. That makes it even less defensible to push unfinished thinking onto someone else.
Measure the recipient’s effort too.
Choose one recurring deliverable and track preparation, review, clarification and rework for four weeks.
Ask whether it actually helped the recipient act.
If AI reduced preparation from four hours to 30 minutes but added three hours of clarification and review for five other people, that is not obviously a productivity gain.
Shorten or stop what adds little value, and recognise people who make the next person’s job easier.
Make the writing standard shared.
Where your tools support it, add organisation-wide writing instructions. For example, Claude Team and Enterprise let owners set shared instructions.
A short starting point:
For workplace writing, ask for the audience, purpose and author’s key points if missing. Preserve their meaning and voice. Keep drafts concise, flag unsupported claims and missing reasoning, and make the recommendation or requested action clear. Do not invent evidence or imply checks were completed.
Test this on real examples.
Shared instructions can guide the tool. They cannot take ownership of the work. The person pressing Send still needs to review it and stand behind it.
Managers need to model this themselves.
Forwarding an unread AI document with “Thoughts?” gives the team another assignment while leaving your own thinking unfinished.
I use AI because it helps me do more. Before pressing Send, I want to ask: have I made this easier for the next person to understand and act on, or just easier for myself to send?
Reader correspondence is welcome at me@xiongzhitao.me.
A note on this note
This note is written in a personal capacity. It is not legal, financial or regulatory advice and should not be relied on as such. The reading is a careful one but errors will exist; please email me with any corrections. See the terms of use for more.