As AI-generated output becomes more abundant, deciding what counts as finished work matters more. A Human Directing perspective on the questions raised by workslop.
A report that once took hours is finished with AI in minutes. The prose is fluent and the structure looks polished. Its author sends it off and leaves work on time.
For the colleague receiving it, however, the work is just beginning: checking figures with unclear sources, adding missing conditions, and correcting conclusions that do not fit the actual situation.
How should we evaluate AI productivity in this scenario?
One person spent less time writing. But if someone else spent more time checking and fixing the result, we need to reassess the benefit to the organization as a whole.
Who gets the time AI saves—and who pays for the time needed to verify its output?
As AI becomes part of everyday work, this question should help shape our standards for collaboration.
The work left behind a polished document
A recent Korea Economic Daily article described workplace friction caused by unverified AI reports and meetings recorded and summarized without participants being informed.
The article uses the term “workslop” for AI-generated material that looks credible but lacks the substance needed to do the job. The recipient must discover and fill the gaps. Read the source article in Korean →
AI gives documents the appearance of completion remarkably quickly. Once a blank page has headings and a conclusion, it is easy to feel that the task is done.
But creating a document and completing the work are different events.
Meeting minutes are not complete simply because the conversation has been shortened. They need to show what was decided, who will do what, and by when. A proposal needs implementation conditions and reasons for its recommendations, alongside persuasive prose.
If we skip those steps because the document looks finished, the unfinished work passes to the next person.
Human Directing includes defining completion
Human Directing can be understood as the ability to guide AI effectively. Yet important work remains after the instruction has been given.
Someone must assess whether the output serves its purpose, check the evidence, and make it usable for its intended recipient.
That starts with a few questions before generation:
- Whose decision should this output help, and what is that decision?
- Which information must be accurate?
- How will confirmed facts be distinguished from assumptions?
- What conditions must be met before the output is handed over?
Without those criteria, the goal becomes “produce a report.” With them, the goal becomes “help someone make the decision they need to make.”
The same AI can produce very different outcomes depending on that framing.
A human director guides the generation process and helps define the conditions under which its output can count as complete.
Productivity includes the next person’s time
Individual time savings are an important measure of AI’s value. In collaborative work, however, we also need to account for time spent after the handoff.
If an author saves two hours but a colleague spends three hours correcting the result, the total workload has increased by one hour, measured in time alone.
Alongside writing speed, we should therefore examine how many follow-up questions recipients need to ask, how much time review and correction take, and whether implementation leads to rework.
A useful deliverable helps the next person understand what they can rely on. It provides evidence, marks assumptions, and makes unresolved decisions visible.
These details may not make a document look more impressive. They do help establish the trust collaboration needs.
Authenticity also requires judgment
The same principle applies to AI-assisted communication.
AI can help someone express gratitude or comfort more clearly. Helping people put feelings into words when they struggle to do so has real value.
But performance reviews, apologies, and conflict resolution require a concrete understanding of another person’s experience. There is work to do before polishing the language.
That work includes considering what the person has experienced, remembering their contributions, and deciding what responsibility we ourselves need to accept.
If those steps are missing, fluent wording may leave the recipient feeling that their situation has not been properly considered.
Meeting transcription similarly requires attention to relationships as well as convenience. Participants need to know how their words will be recorded, who will receive the record, and how it will be used.
The person using AI needs to consider both the recipient of the output and the people whose words or experiences become its input.
Where might the advantages of AI use shift next?
I see three possible developments. These are hypotheses from a Human Directing perspective, rather than established predictions reported by the source article.
1. Usability will matter alongside output volume
As the ability to generate documents quickly becomes widespread, volume alone will become a less useful way to distinguish performance.
Whether a document supports a decision, reduces correction and rework, or leads to effective action is likely to become more important.
2. Verification will become part of workflow design
As output grows, asking people to reread everything from scratch becomes difficult to sustain.
Checks for calculation errors, missing fields, and inconsistencies between sources can be automated. Decisions with greater consequences can be routed to an accountable reviewer.
Adding a second AI as a reviewer is not sufficient on its own. A workflow also needs comparison sources, conditions that trigger a pause, and a clear allocation of final decision authority.
3. Domain knowledge will become a more valuable basis for directing AI
As prose becomes more fluent, recognizing gaps in substance becomes more important.
Knowing what is missing, which assumption does not fit reality, and where a problem may emerge often draws on experience and context within a field.
People who can explain that experience and turn it into usable criteria can give AI a more specific direction. Their expertise can become a resource that colleagues and systems can use.
What if we spent some of the saved time on judgment?
Finishing work earlier with AI is a welcome improvement. While enjoying that benefit, we should also examine where any remaining burden has gone.
Has it become a colleague’s additional review, a customer’s confusion, or rework during implementation?
If we invest some of the time AI saves in better questions and focused checks, individual time savings may translate into time savings for the people working with us.
Human Directing includes deciding where that time should go: clarifying purpose, arranging appropriate verification, and taking responsibility for the handoff.
After finishing a task with AI, we might ask one final question:
When my work got faster, did the next person’s work become easier too?
Research note: Human Directing is a provisional concept for exploring how human roles, judgment, and intervention change as AI advances. The outlook and role descriptions in this essay are hypotheses to test and revise, not claims about abilities that must remain exclusively human forever.
Source: The Korea Economic Daily, “AI helped me leave work on time…”: Why workers are frustrated (Korean article; title translated for reference), October 13, 2026.
Previous research note: If AI Decides When Humans Should Intervene, What Should Humans Decide? →
답글 남기기